Skip to content

#1 Persistent memory for AI coding agents based on real-world benchmarks

fix: update doctor diagnostics and 0.9.31 documentation (#895)

Rohit Ghumareb2629cb1e709
diff --git a/README.md b/README.md
index 5f291b54..07d9286b 100644
--- a/README.md
+++ b/README.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/stat-recall.svg"><img src="assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/stat-tokens.svg"><img src="assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/stat-tools.svg"><img src="assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/stat-hooks.svg"><img src="assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/stat-deps.svg"><img src="assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -334,225 +332,47 @@ Latest release notes: [CHANGELOG.md](CHANGELOG.md).
---
-<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-benchmarks.svg"><img src="assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval Accuracy
-
-**coding-agent-life-v1** (in-house corpus, sandbox-reproducible)
+### What's new in 0.9.31
-| Adapter | P@5 | R@5 | Top-5 hit rate | p50 latency |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% top-5 hit rate at the **P@5 math ceiling** for this corpus (0.240, see scorecard). Hybrid retrieves every gold session; grep misses 1 of 2 gold on the multi-session temporal query. Lift is **recall + temporal**, not aggregate precision. This benchmark is small and gold-sparse; the larger LongMemEval-S below differentiates better. Full per-type breakdown + correction note: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 questions)
-
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
+- **More focused recall:** project and agent scopes, saved-memory retrieval, explicit context budgets and full-source expansion.
+- **Less repeated work:** incremental session summaries skip unchanged observations; exports page through stored data.
+- **More native integrations:** installable Pi package, OMP and DeepSeek Harness plugins, and improved Cursor workspace attribution.
+- **Safer recovery:** compression, import validation and runtime recovery fixes, with status and activity visible in the viewer.
-### Token Savings
+[Release notes and contributor credits](https://github.com/rohitg00/agentmemory/releases/tag/v0.9.31).
-| Approach | Tokens/yr | Cost/yr |
-|---|---|---|
-| Paste full context | 19.5M+ | Impossible (exceeds window) |
-| LLM-summarized | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + local embeddings | ~170K | **$0** |
+<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-benchmarks.svg"><img src="assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-</td>
-</tr>
-</table>
+**Useful recall without a model API.** On LongMemEval-S, AgentMemory's keyless BM25 component found at least one relevant source for **96.2%** of answerable questions and **every required source for 84.7%**.
-> Embedding model: `all-MiniLM-L6-v2` (local, free, no API key). Full reports: [`benchmark/LONGMEMEVAL.md`](benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](benchmark/QUALITY.md), [`benchmark/SCALE.md`](benchmark/SCALE.md). Competitor comparison: [`benchmark/COMPARISON.md`](benchmark/COMPARISON.md) covering agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
+| LongMemEval-S, top 5 distinct sessions | AgentMemory BM25 | Lexical baseline |
+|---|---:|---:|
+| Every required source found (strict recall-all@5) | **84.7%** | 72.1% |
+| At least one relevant source found (any-hit@5) | **96.2%** | 88.7% |
-**Reproduce locally:** [`eval/README.md`](eval/README.md), an adapter-pluggable harness for LongMemEval `_s` (public 500-Q) + `coding-agent-life-v1` (in-house 15-session corpus). Grep / vector / agentmemory adapters score side-by-side, NDJSON output, published scorecards land in [`docs/benchmarks/`](docs/benchmarks/).
+That is **12.6 percentage points higher complete-source retrieval** than the lexical baseline on the same inputs, with no embedding model or paid API. Measured on 2026-10-28: 500 questions per adapter, 470 answerable, 30 excluded from retrieval scoring, zero execution failures.
-**Pairs with [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), and [Graphify](https://github.com/safishamsi/graphify).** Code-graph indexing, multi-agent build pipelines, and broader knowledge graphs across docs / PDFs / images / videos. agentmemory remembers the work; those three projects light up the rest of the context layer. Recipes + question-routing table: [`docs/recipes/pairings.md`](docs/recipes/pairings.md).
+This component test measures retrieval from supplied sessions. Full-system answer quality, hybrid retrieval and competitor rankings require separate evaluations. [Read the complete results, including the smaller coding fixture, limitations and reproduction steps](docs/benchmarks/2026-10-28-retrieval-evaluation.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-competitors.svg"><img src="assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Built-in (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Type</strong></td>
-<td>Memory engine + MCP server</td>
-<td>Memory layer API</td>
-<td>Full agent runtime</td>
-<td>Personal AI</td>
-<td>Memory API + app</td>
-<td>Team memory hub (LLM proxy)</td>
-<td>Vector memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Memory system</td>
-<td>Static file</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Self-reported</td>
-<td>PersonaMem 76% (self-reported)</td>
-<td>~96.6% (self-reported)</td>
-<td>94.4% (self-reported)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hooks (zero manual effort)</td>
-<td>Manual <code>add()</code> calls</td>
-<td>Agent self-edits</td>
-<td>Manual</td>
-<td>API-side extraction</td>
-<td>Proxy interception (base-URL swap)</td>
-<td>Manual</td>
-<td>API extraction</td>
-<td>Manual</td>
-<td>Manual editing</td>
-</tr>
-<tr>
-<td><strong>Search</strong></td>
-<td>BM25 + Vector + Graph (RRF fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semantic</td>
-<td>Vector + RAG</td>
-<td>4 asset types (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vector-only</td>
-<td>Vector + semantic</td>
-<td>Decay-weighted</td>
-<td>Loads everything into context</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (no coordination)</td>
-<td>Within Letta runtime only</td>
-<td>No</td>
-<td>No</td>
-<td>Team roles + shared assets</td>
-<td>No</td>
-<td>Scoped only</td>
-<td>Multi-agent shared</td>
-<td>Per-agent files</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>None (any MCP client)</td>
-<td>None</td>
-<td>High (must use Letta)</td>
-<td>Standalone</td>
-<td>None</td>
-<td>Proxy fronts every model call</td>
-<td>None</td>
-<td>Oracle Database</td>
-<td>None</td>
-<td>Per-agent format</td>
-</tr>
-<tr>
-<td><strong>External deps</strong></td>
-<td>No separate database service required</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>Multiple</td>
-<td>Managed cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>None</td>
-<td>None</td>
-</tr>
-<tr>
-<td><strong>Memory lifecycle</strong></td>
-<td>4-tier consolidation + decay + auto-forget</td>
-<td>Passive extraction</td>
-<td>Agent-managed</td>
-<td>Manual</td>
-<td>Auto-forget</td>
-<td>Manual review; auto-routing in progress</td>
-<td>None</td>
-<td>Not stated</td>
-<td>Decay + consolidation</td>
-<td>Manual pruning</td>
-</tr>
-<tr>
-<td><strong>Token efficiency</strong></td>
-<td>~1,900 tokens/session ($10/yr)</td>
-<td>Varies by integration</td>
-<td>Core memory in context</td>
-<td>Varies</td>
-<td>Cloud pricing</td>
-<td>Not stated</td>
-<td>No token budget</td>
-<td>LLM-backed (varies)</td>
-<td>Varies</td>
-<td>22K+ tokens at 240 obs</td>
-</tr>
-<tr>
-<td><strong>Real-time viewer</strong></td>
-<td>Yes (port 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Yes (default)</td>
-<td>Optional</td>
-<td>Optional</td>
-<td>Yes</td>
-<td>No (cloud-only)</td>
-<td>Yes (Docker)</td>
-<td>Yes</td>
-<td>Yes (Oracle DB)</td>
-<td>Yes</td>
-<td>Yes</td>
-</tr>
-</table>
+**AgentMemory is built for the coding workflow you already have.** Capture supported agent activity automatically, share project history across tools, retrieve scoped memories, and inspect the original sources in a local viewer. Lessons, actions, versioned memories and import/export ship in the Apache-2.0 project; keyword recall needs no account or model API key.
-<sub>Benchmark note: only agentmemory's R@5 is our own measured result (LongMemEval-S, reproducible from <a href="benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). The mem0 and Letta figures are their published LoCoMo numbers (a different dataset); the MemPalace, supermemory, TencentDB (PersonaMem), and oracleagentmemory figures are vendor self-reported claims we have not independently reproduced (oracleagentmemory's run used GPT-5.5 against an Oracle AI Database). Shown side by side for ballpark only, not a head-to-head on identical data. Star counts are approximate and drift over time.</sub>
+Choose by workload and deployment. Other products also offer local deployment or agent plugins, so those alone are not useful differentiators.
-**Newer entrants** worth knowing, compared in depth in [`benchmark/COMPARISON.md`](benchmark/COMPARISON.md):
-
-| System | ⭐ | Angle |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal knowledge graph; strongest published temporal-query results (LongMemEval 63.8%), but graph builds asynchronously so fresh facts can lag |
-| Cognee | 30K | Document-to-knowledge-graph ingestion, Python-only, built for structured entity extraction rather than session capture |
+| Project | Best starting point for |
+|---|---|
+| **AgentMemory** | Shared local memory for existing coding agents, automatic capture, scoped recall and a live viewer |
+| [Mem0](https://docs.mem0.ai/introduction) | Memory APIs, hosted or self-hosted, with SDKs and agent plugins |
+| [Supermemory](https://supermemory.ai/docs/self-hosting/overview) | Memory and document context through a hosted API or local server |
+| [Zep / Graphiti](https://github.com/getzep/graphiti) | Enterprise context and temporal knowledge graphs |
+| [Letta](https://docs.letta.com/) | Building stateful agents with their own managed memory |
+| [Cognee](https://github.com/topoteretes/cognee) | Turning documents, code and conversations into searchable knowledge |
+| [claude-mem](https://github.com/thedotmack/claude-mem) | Capturing and summarizing coding sessions across supported agents |
-AgentMemory combines coding-agent capture, searchable project history and a local viewer, with keyless operation available from the first run.
+[Detailed comparison and sources](benchmark/COMPARISON.md), checked 2026-10-28. Deployment options, provider costs and integrations vary; scores from different datasets and metrics are not a head-to-head ranking.
---
@@ -787,9 +607,9 @@ Full guide: [`integrations/hermes/`](integrations/hermes/)
Start the memory server: `npx -y @agentmemory/agentmemory@latest`
-#### Native skills via `npx skills add` (50+ agents)
+#### Native skills via `npx skills add`
-agentmemory ships 17 skills in the Claude-Code-style `<dir>/SKILL.md` format: 9 invocable action skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) and 8 reference skills the agent loads on demand (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). The reference skills carry data tables generated from source, so they never drift. The [`skills`](https://npmjs.com/package/skills) CLI by vercel-labs auto-installs them into the calling agent's native skill directory across 50+ agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, and more):
+agentmemory ships 17 skills in the Claude-Code-style `<dir>/SKILL.md` format: 9 invocable action skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) and 8 reference skills the agent loads on demand (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). The reference skills carry data tables generated from source, so they never drift. The [`skills`](https://npmjs.com/package/skills) CLI by vercel-labs auto-installs them into the calling agent's native skill directory across supported agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, and more):
```bash
npx skills add rohitg00/agentmemory -y # auto-detects the calling agent
@@ -845,7 +665,7 @@ The agentmemory entry is the **same MCP server block** across every host that us
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installs MCP and capture hooks in the shared customization directory. See [Antigravity setup and limits](docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` uses the same MCP and hook configuration as current IDE versions. Existing installations should refresh with `--force`; see the [upgrade notes](docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` writes the user-level config. Workspace overrides go in `.kiro/settings/mcp.json` next to your code. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` writes the standard `mcpServers` block. Warp also auto-discovers skills from `.claude/skills/`; once the Claude Code plugin is installed the 8 agentmemory skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) appear natively in Warp's slash-command palette. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` writes the standard `mcpServers` block. Warp also auto-discovers skills from `.claude/skills/`; once the Claude Code plugin is installed the 9 agentmemory skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) appear natively in Warp's slash-command palette. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` writes the standard `mcpServers` block. VS Code extension users: paste the same block via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferred) or `config.json` (legacy) | `agentmemory connect continue` creates `config.yaml` from scratch when neither exists, or modifies existing `config.json`. **If you already have `config.yaml`** the adapter prints the exact block to paste under `mcpServers:`; it won't silently rewrite your yaml so your existing comments and anchors are preserved. Continue uses array form (not object) for `mcpServers`. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` writes under `context_servers` (Zed's key, NOT `mcpServers`). Remote MCP servers can be wired via `{"url": "..."}` instead. |
@@ -938,15 +758,15 @@ Session 2: "Now add rate limiting"
### vs built-in agent memory
-Every AI coding agent ships with built-in memory: Claude Code has `MEMORY.md`, Cursor has notepads, Cline has memory bank. These work like sticky notes. agentmemory is the searchable database behind the sticky notes.
+Agent instructions and memory files remain useful for stable project guidance. AgentMemory adds searchable session history, source-linked memories and coordination across connected agents.
| | Built-in (CLAUDE.md) | agentmemory |
|---|---|---|
| Upkeep | Manually curated instructions | Indexed session history and memories |
| Search | Loads everything into context | BM25 + vector + graph (top-K only) |
-| Token cost | 22K+ at 240 observations | ~1,900 tokens (92% less) |
+| Context budget | Depends on the files included | Configurable recall budget with source expansion |
| Cross-agent | Per-agent files | MCP + REST (any agent) |
-| Coordination | None | Leases, signals, actions, routines |
+| Coordination | Depends on the agent | Leases, signals, actions, routines |
| Observability | Read files manually | Real-time viewer on :3113 |
---
@@ -957,7 +777,7 @@ Every AI coding agent ships with built-in memory: Claude Code has `MEMORY.md`, C
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID, durable capture inbox, offline spool when unavailable
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -966,7 +786,7 @@ PostToolUse hook fires
-> Index in BM25, plus vectors when enabled
Stop / SessionEnd hook fires
- -> Summarize session
+ -> Incremental summary when new observations are available
-> Knowledge graph extraction (if GRAPH_EXTRACTION_ENABLED=true)
-> Slot reflection (if AGENTMEMORY_SLOTS=true and AGENTMEMORY_REFLECT=true)
@@ -1046,7 +866,7 @@ Fused with Reciprocal Rank Fusion (RRF, k=60) and session-diversified (max 3 res
When a vector index is populated, `mem::search` (behind `memory_recall`) uses the hybrid BM25 + vector ranker. Without embeddings it uses BM25. `smart-search` can additionally fuse structural graph matches when graph data exists, including in keyless mode. Lesson recall runs on a dedicated in-memory BM25 index instead of scanning the whole corpus per query. Superseded memory versions are excluded from every recall path; the version chain keeps their history.
-Vectors survive a crash or force-kill. The vector index is saved in buckets at most every `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutes). Every vector added or removed in between is also written right away to a small pending log in the state store, and the next start replays it without calling the embedding provider. Each successful save empties the log. Documents that still have no vector after the replay are re-embedded in the background in batches of `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) until none are left, and a backfill that is stopped continues at the next start. `/agentmemory/status` and the viewer show the pending log size and the backfill state. Keyless installs write nothing.
+Vectors survive a crash or force-kill. The vector index is saved in buckets at most every `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutes). Every vector added or removed in between is also written right away to a small pending log in the state store, and the next start replays it without calling the embedding provider. Each successful save empties the log. For older documents without vectors, `AGENTMEMORY_VECTOR_BACKFILL=all` opts into a bounded backfill at startup, capped by `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500). Keep the opt-in set across restarts to process more documents. See [source and vector recovery](docs/reliability/source-and-vector-recovery.md). `/agentmemory/status` and the viewer show the pending log size and the backfill state. Keyless installs write nothing.
BM25 tokenizes Greek, Cyrillic, Hebrew, Arabic, and accented Latin out of the box. For Chinese / Japanese / Korean memories, install the optional segmenters (`npm install @node-rs/jieba tiny-segmenter`) to split CJK runs into word-level tokens; without them, agentmemory soft-falls to whole-run tokenization and prints a one-time hint on stderr.
@@ -1062,9 +882,9 @@ The normal npm install includes the optional `@huggingface/transformers` runtime
| Provider | Model | Cost | Notes |
|---|---|---|---|
-| **Local (recommended opt-in)** | `all-MiniLM-L6-v2` | Free | On-device after the first model download, +8pp recall over BM25-only |
+| **Local (recommended opt-in)** | `all-MiniLM-L6-v2` | Free | On-device after the first model download |
| Gemini | `gemini-embedding-001` | Free tier | 100+ languages, 768/1536/3072 dims (MRL), 2048-token input. Replaces `text-embedding-004` ([deprecated, shutdown Jan 14, 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Highest quality |
+| OpenAI | `text-embedding-3-small` | [Provider pricing](https://openai.com/api/pricing/) | Remote embeddings |
| Voyage AI | `voyage-code-3` | Paid | Optimized for code |
| Cohere | `embed-english-v3.0` | Free trial | General purpose |
| OpenRouter | Any model | Varies | Multi-model proxy |
diff --git a/READMEs/README.ar-SA.md b/READMEs/README.ar-SA.md
index c353c033..b1361afe 100644
--- a/READMEs/README.ar-SA.md
+++ b/READMEs/README.ar-SA.md
@@ -68,8 +68,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -335,223 +333,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### دقة الاسترجاع
-
-**coding-agent-life-v1** (مجموعة بيانات داخلية، قابلة لإعادة الإنتاج في بيئة معزولة)
-
-| المحوّل | P@5 | R@5 | معدل الإصابة ضمن أفضل 5 | زمن الاستجابة p50 |
-|---|---|---|---|---|
-| **agentmemory الهجين** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| خط أساس grep | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-معدل إصابة 100% ضمن أفضل 5 عند **سقف P@5 الرياضي** لهذه المجموعة (0.240، راجع بطاقة النتائج). يسترجع النظام الهجين كل جلسة مرجعية (gold)؛ بينما يفوّت grep جلسة واحدة من جلستين مرجعيتين في الاستعلام الزمني متعدد الجلسات. التحسّن هنا في **الاسترجاع + البعد الزمني**، لا في الدقة الكلية. هذا المقياس صغير وقليل الجلسات المرجعية؛ ويميّز المقياس الأكبر LongMemEval-S أدناه بشكل أفضل. التفصيل الكامل حسب النوع وملاحظة التصحيح: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025، 500 سؤال)
+نتائج الاسترجاع الحالية، وتعريف المقاييس، وخطوات إعادة الاختبار:
-| النظام | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25 فقط (خيار احتياطي) | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### توفير الرموز (Tokens)
-
-| الأسلوب | الرموز/سنويًا | التكلفة/سنويًا |
-|---|---|---|
-| لصق السياق الكامل | 19.5M+ | غير ممكن (يتجاوز النافذة) |
-| ملخَّص بواسطة LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + تضمينات محلية | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> نموذج التضمين: `all-MiniLM-L6-v2` (محلي، مجاني، بلا مفتاح API). التقارير الكاملة: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md)، [`benchmark/QUALITY.md`](../benchmark/QUALITY.md)، [`benchmark/SCALE.md`](../benchmark/SCALE.md). مقارنة بالمنافسين: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) وتشمل agentmemory في مواجهة mem0 وLetta وKhoj وsupermemory وTencentDB Agent Memory وMemPalace وZep/Graphiti وCognee وHippo.
-
-**إعادة الإنتاج محليًا:** [`eval/README.md`](../eval/README.md)، وهو إطار اختبار (harness) قابل لتوصيل المحوّلات لـ LongMemEval `_s` (500 سؤال علنية) + `coding-agent-life-v1` (مجموعة داخلية من 15 جلسة). تُقيَّم محوّلات grep / المتجهات / agentmemory جنبًا إلى جنب، بمخرجات NDJSON، وتُنشر بطاقات النتائج في [`docs/benchmarks/`](../docs/benchmarks/).
-
-**يتكامل مع [codegraph](https://github.com/colbymchenry/codegraph) و[Understand Anything](https://github.com/Lum1104/Understand-Anything) و[Graphify](https://github.com/safishamsi/graphify).** فهرسة الرسم البياني للشفرة (code-graph)، وخطوط بناء متعددة الوكلاء، ورسوم بيانية معرفية أوسع تمتد عبر المستندات/ملفات PDF/الصور/الفيديوهات. يتذكّر agentmemory العمل؛ وتُضيء هذه المشاريع الثلاثة بقية طبقة السياق. الوصفات وجدول توجيه الأسئلة: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>مدمج (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>النوع</strong></td>
-<td>محرك ذاكرة + خادم MCP</td>
-<td>واجهة برمجية لطبقة الذاكرة</td>
-<td>بيئة تشغيل كاملة للوكيل</td>
-<td>ذكاء اصطناعي شخصي</td>
-<td>واجهة برمجية للذاكرة + تطبيق</td>
-<td>محور ذاكرة جماعي (وكيل LLM وسيط)</td>
-<td>ذاكرة متجهية (مفتوحة المصدر)</td>
-<td>محرك ذاكرة (قاعدة بيانات Oracle)</td>
-<td>نظام ذاكرة</td>
-<td>ملف ثابت</td>
-</tr>
-<tr>
-<td><strong>دقة الاسترجاع R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>غير متوفر</td>
-<td>مُعلَن من الشركة نفسها</td>
-<td>PersonaMem 76% (مُعلَن من الشركة نفسها)</td>
-<td>~96.6% (مُعلَن من الشركة نفسها)</td>
-<td>94.4% (مُعلَن من الشركة نفسها)</td>
-<td>غير متوفر</td>
-<td>غير متوفر (grep)</td>
-</tr>
-<tr>
-<td><strong>الالتقاط التلقائي</strong></td>
-<td>12 خطافًا (بلا أي جهد يدوي)</td>
-<td>استدعاءات <code>add()</code> يدوية</td>
-<td>تعديلات ذاتية من الوكيل</td>
-<td>يدوي</td>
-<td>استخراج من جانب الواجهة البرمجية</td>
-<td>اعتراض عبر وكيل وسيط (باستبدال base-URL)</td>
-<td>يدوي</td>
-<td>استخراج عبر API</td>
-<td>يدوي</td>
-<td>تعديل يدوي</td>
-</tr>
-<tr>
-<td><strong>البحث</strong></td>
-<td>BM25 + المتجهات + الرسم البياني (دمج RRF)</td>
-<td>المتجهات + الرسم البياني</td>
-<td>متجهات (أرشيفية)</td>
-<td>دلالي</td>
-<td>متجهات + RAG</td>
-<td>4 أنواع من الأصول (محادثة / مهارة / Wiki / CodeGraph)</td>
-<td>متجهات فقط</td>
-<td>متجهات + دلالي</td>
-<td>مرجَّح بالتلاشي</td>
-<td>يحمّل كل شيء إلى السياق</td>
-</tr>
-<tr>
-<td><strong>تعدد الوكلاء</strong></td>
-<td>MCP + REST + حجوزات (leases) + إشارات</td>
-<td>API (بلا تنسيق)</td>
-<td>داخل بيئة تشغيل Letta فقط</td>
-<td>لا</td>
-<td>لا</td>
-<td>أدوار جماعية + أصول مشتركة</td>
-<td>لا</td>
-<td>محدود النطاق فقط</td>
-<td>مشترك بين وكلاء متعددين</td>
-<td>ملفات لكل وكيل على حدة</td>
-</tr>
-<tr>
-<td><strong>الارتهان لإطار عمل محدد</strong></td>
-<td>لا يوجد (أي عميل MCP)</td>
-<td>لا يوجد</td>
-<td>مرتفع (يجب استخدام Letta)</td>
-<td>مستقل</td>
-<td>لا يوجد</td>
-<td>وكيل وسيط يتقدّم كل استدعاء للنموذج</td>
-<td>لا يوجد</td>
-<td>Oracle Database</td>
-<td>لا يوجد</td>
-<td>تنسيق خاص بكل وكيل</td>
-</tr>
-<tr>
-<td><strong>الاعتماديات الخارجية</strong></td>
-<td>لا يتطلب خدمة قاعدة بيانات منفصلة</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + قاعدة بيانات متجهية</td>
-<td>متعددة</td>
-<td>سحابة مُدارة</td>
-<td>حزمة Docker (Core + Hub + Proxy)</td>
-<td>مخزن متجهات</td>
-<td>Oracle AI Database</td>
-<td>لا يوجد</td>
-<td>لا يوجد</td>
-</tr>
-<tr>
-<td><strong>دورة حياة الذاكرة</strong></td>
-<td>توحيد رباعي المستويات + تلاشي + نسيان تلقائي</td>
-<td>استخراج سلبي (passive)</td>
-<td>يديرها الوكيل</td>
-<td>يدوي</td>
-<td>نسيان تلقائي</td>
-<td>مراجعة يدوية؛ والتوجيه التلقائي قيد التطوير</td>
-<td>لا يوجد</td>
-<td>غير مُحدَّد</td>
-<td>تلاشي + توحيد</td>
-<td>تقليم يدوي</td>
-</tr>
-<tr>
-<td><strong>كفاءة استخدام الرموز</strong></td>
-<td>~1,900 رمز/جلسة (10$/سنويًا)</td>
-<td>يتفاوت حسب التكامل</td>
-<td>الذاكرة الأساسية داخل السياق</td>
-<td>متفاوت</td>
-<td>تسعير سحابي</td>
-<td>غير مُحدَّد</td>
-<td>بلا حد لعدد الرموز</td>
-<td>معتمد على LLM (متفاوت)</td>
-<td>متفاوت</td>
-<td>أكثر من 22K رمز عند 240 ملاحظة</td>
-</tr>
-<tr>
-<td><strong>العارض اللحظي</strong></td>
-<td>نعم (المنفذ 3113)</td>
-<td>لوحة تحكم سحابية</td>
-<td>لوحة تحكم سحابية</td>
-<td>واجهة ويب</td>
-<td>لوحة تحكم سحابية</td>
-<td>واجهة ويب للمحور (Hub)</td>
-<td>لا</td>
-<td>لا</td>
-<td>لا</td>
-<td>لا</td>
-</tr>
-<tr>
-<td><strong>استضافة ذاتية</strong></td>
-<td>نعم (افتراضيًا)</td>
-<td>اختياري</td>
-<td>اختياري</td>
-<td>نعم</td>
-<td>لا (سحابي فقط)</td>
-<td>نعم (Docker)</td>
-<td>نعم</td>
-<td>نعم (Oracle DB)</td>
-<td>نعم</td>
-<td>نعم</td>
-</tr>
-</table>
-
-<sub>ملاحظة على المقياس: نتيجة R@5 الخاصة بـ agentmemory فقط هي قياسنا الفعلي (LongMemEval-S، قابلة لإعادة الإنتاج من <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). أما أرقام mem0 وLetta فهي أرقامهما المنشورة على LoCoMo (مجموعة بيانات مختلفة)؛ وأرقام MemPalace وsupermemory وTencentDB (PersonaMem) وoracleagentmemory هي ادعاءات مُعلَنة من الشركات نفسها لم نُعِد إنتاجها بشكل مستقل (استخدمت عملية تشغيل oracleagentmemory نموذج GPT-5.5 مقابل قاعدة بيانات Oracle AI Database). عُرضت جنبًا إلى جنب للمقارنة التقريبية فقط، لا كمواجهة مباشرة على بيانات متطابقة. أعداد النجوم تقريبية وتتغير مع الوقت.</sub>
-
-**داخلون جدد** تستحق المعرفة، مع مقارنة تفصيلية في [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| النظام | ⭐ | الزاوية |
-|--------|---|-------|
-| Zep / Graphiti | 30K | رسم بياني معرفي زمني؛ يملك أقوى نتائج منشورة للاستعلامات الزمنية (LongMemEval 63.8%)، لكن بناء الرسم البياني يتم بشكل غير متزامن، فقد تتأخر الحقائق الجديدة |
-| Cognee | 30K | استيعاب يحوّل المستندات إلى رسم بياني معرفي، مخصص لـ Python فقط، ومبني لاستخراج الكيانات البنيوية لا لالتقاط الجلسات |
+مقارنة محدثة بين المنتجات وخيارات النشر، مع المصادر:
-يجمع AgentMemory التقاط الجلسات وسجل المشروع القابل للبحث والعارض المحلي، مع العمل دون مفتاح منذ التشغيل الأول.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -846,7 +638,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | يُثبِّت `agentmemory connect antigravity --with-hooks` خطافات MCP والالتقاط في دليل التخصيص (customization) المشترك. راجع [إعداد Antigravity وحدوده](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | يستخدم `agentmemory connect antigravity-cli --with-hooks` نفس إعدادات MCP والخطافات المستخدمة في إصدارات IDE الحالية. ينبغي للتثبيتات الحالية التحديث باستخدام `--force`؛ راجع [ملاحظات الترقية](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | يكتب `agentmemory connect kiro` الإعدادات على مستوى المستخدم. وتوضع تجاوزات مساحة العمل في `.kiro/settings/mcp.json` بجانب شفرتك. |
-| **Warp** | `~/.warp/.mcp.json` | يكتب `agentmemory connect warp` كتلة `mcpServers` القياسية. يكتشف Warp أيضًا المهارات تلقائيًا من `.claude/skills/`؛ وبمجرد تثبيت إضافة Claude Code، تظهر مهارات agentmemory الثماني (`remember` و`recall` و`recap` و`handoff` و`forget` و`commit-context` و`commit-history` و`session-history`) أصليًا في لوحة أوامر الشرطة في Warp. |
+| **Warp** | `~/.warp/.mcp.json` | يكتب `agentmemory connect warp` كتلة `mcpServers` القياسية. يكتشف Warp أيضًا المهارات تلقائيًا من `.claude/skills/`؛ وبمجرد تثبيت إضافة Claude Code، تظهر مهارات agentmemory التسع (`remember` و`recall` و`recap` و`handoff` و`forget` و`lesson` و`commit-context` و`commit-history` و`session-history`) أصليًا في لوحة أوامر الشرطة في Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | يكتب `agentmemory connect cline` كتلة `mcpServers` القياسية. لمستخدمي إضافة VS Code: الصق الكتلة نفسها عبر Cline Settings ← MCP Servers ← Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (مفضّل) أو `config.json` (قديم) | ينشئ `agentmemory connect continue` ملف `config.yaml` عند غياب الإعدادات، أو يعدّل `config.json` الموجود. بالنسبة إلى YAML موجود، يعرض الكتلة لإضافتها تحت `mcpServers:` للحفاظ على التعليقات والمراسي. يستخدم Continue مصفوفة لا كائنًا. |
| **Zed** | `~/.config/zed/settings.json` | يكتب `agentmemory connect zed` تحت `context_servers` (مفتاح Zed الخاص، وليس `mcpServers`). يمكن ربط خوادم MCP البعيدة بدلًا من ذلك عبر `{"url": "..."}`. |
@@ -928,7 +720,6 @@ npx -y @agentmemory/mcp
|---|---|---|
| الصيانة | تعليمات تُدار يدويًا | سجل جلسات وذكريات مفهرسة |
| البحث | يحمّل كل شيء إلى السياق | BM25 + متجهات + رسم بياني (أفضل K فقط) |
-| تكلفة الرموز | أكثر من 22K عند 240 ملاحظة | ~1,900 رمز (أقل بنسبة 92%) |
| بين الوكلاء | ملفات لكل وكيل | MCP + REST (أي وكيل) |
| التنسيق | لا يوجد | حجوزات، وإشارات، وإجراءات، ومسارات عمل (routines) |
| قابلية الرصد | قراءة الملفات يدويًا | عارض لحظي على :3113 |
@@ -941,7 +732,7 @@ npx -y @agentmemory/mcp
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1026,7 +817,7 @@ SessionStart hook fires
عندما يكون فهرس المتجهات مُعبَّأً، يستخدم `mem::search` (الذي يقف خلف `memory_recall`) المُصنِّف الهجين BM25 + المتجهات. وبدون تضمينات يستخدم BM25. ويمكن لـ `smart-search` أيضًا دمج تطابقات الرسم البياني البنيوية عندما تتوفر بيانات رسم بياني، حتى في وضع عدم استخدام المفاتيح. ويعمل استرجاع الدروس المستفادة على فهرس BM25 مخصص في الذاكرة بدلًا من مسح المجموعة الكاملة في كل استعلام. وتُستبعد نسخ الذاكرة المُستبدَلة من كل مسار استرجاع؛ وتحتفظ سلسلة الإصدارات بتاريخها.
-تنجو المتجهات من الانهيار أو الإيقاف القسري. ويُحفظ فهرس المتجهات في حُزَم مرة واحدة على الأكثر كل `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 دقائق). ويُكتب أيضًا كل متجه يُضاف أو يُحذف بين ذلك فورًا في سجل انتظار صغير في مخزن الحالة، ويُعاد تشغيله عند البدء التالي دون استدعاء مزوّد التضمين. ويُفرَّغ السجل عند كل حفظ ناجح. وتُعاد تضمين المستندات التي لا تزال بلا متجه بعد إعادة التشغيل في الخلفية على دفعات بحجم `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) إلى أن لا يبقى منها شيء، ويستكمل أي ملء رجعي (backfill) متوقف في البدء التالي. يُظهر `/agentmemory/status` والعارض حجم سجل الانتظار وحالة الملء الرجعي. ولا تكتب التثبيتات بلا مفاتيح أي شيء.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
يُجزِّئ BM25 إلى رموز (tokens) اليونانية والسيريلية والعبرية والعربية واللاتينية المُشكَّلة جاهزًا دون أي إعداد. وبالنسبة لذكريات الصينية/اليابانية/الكورية، ثبّت المُجزِّئات (segmenters) الاختيارية (`npm install @node-rs/jieba tiny-segmenter`) لتقسيم سلاسل CJK إلى رموز على مستوى الكلمة؛ وبدونها، يتراجع agentmemory بسلاسة إلى تجزيء السلسلة كاملةً ويطبع تلميحًا لمرة واحدة على stderr.
@@ -1042,9 +833,9 @@ EMBEDDING_PROVIDER=local
| المزوّد | النموذج | التكلفة | ملاحظات |
|---|---|---|---|
-| **محلي (مُوصى بتفعيله)** | `all-MiniLM-L6-v2` | مجاني | على الجهاز نفسه بعد أول تنزيل للنموذج، وبزيادة 8 نقاط نسبة مئوية في الاسترجاع عن BM25 فقط |
+| **محلي (مُوصى بتفعيله)** | `all-MiniLM-L6-v2` | مجاني | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | مستوى مجاني | أكثر من 100 لغة، أبعاد 768/1536/3072 (MRL)، ومدخل حتى 2048 رمزًا. يحل محل `text-embedding-004` ([مُستبعَد، يُغلق في 14 يناير 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | 0.02$/مليون | أعلى جودة |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | مدفوع | مُحسَّن للشفرة |
| Cohere | `embed-english-v3.0` | نسخة تجريبية مجانية | للاستخدام العام |
| OpenRouter | أي نموذج | متفاوت | وكيل متعدد النماذج |
diff --git a/READMEs/README.bn-BD.md b/READMEs/README.bn-BD.md
index 4346dd2b..f4844612 100644
--- a/READMEs/README.bn-BD.md
+++ b/READMEs/README.bn-BD.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% রিট্রিভাল R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% কম টোকেন" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP টুল" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 অটো হুক" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 এক্সটার্নাল ডিবি" height="38" /></picture>
@@ -331,223 +329,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="বেঞ্চমার্ক" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### রিট্রিভাল অ্যাকুরেসি
-
-**coding-agent-life-v1** (ইন-হাউস করপাস, স্যান্ডবক্সে রিপ্রোডিউসযোগ্য)
-
-| অ্যাডাপ্টার | P@5 | R@5 | টপ-5 হিট রেট | p50 লেটেন্সি |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep বেসলাইন | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-এই করপাসের জন্য **P@5-এর গাণিতিক সর্বোচ্চ সীমায় (0.240, স্কোরকার্ড দেখুন)** 100% টপ-5 হিট রেট। Hybrid প্রতিটি গোল্ড সেশন রিট্রিভ করে; multi-session temporal কোয়েরিতে grep 2টির মধ্যে 1টি গোল্ড মিস করে। উন্নতিটা **রিকল + টেম্পোরাল**-এর, অ্যাগ্রিগেট প্রিসিশনের নয়। এই বেঞ্চমার্কটি ছোট এবং গোল্ড-স্পার্স; নিচের বড় LongMemEval-S আরও ভালোভাবে পার্থক্য দেখায়। প্রতি-টাইপ সম্পূর্ণ ব্রেকডাউন + সংশোধন নোট: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)।
-
-**LongMemEval-S** (ICLR 2025, 500 প্রশ্ন)
+বর্তমান পুনরুদ্ধারের ফলাফল, মেট্রিকের সংজ্ঞা ও পুনরায় পরীক্ষার ধাপ:
-| সিস্টেম | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| শুধু BM25 ফলব্যাক | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### টোকেন সাশ্রয়
-
-| পদ্ধতি | টোকেন/বছর | খরচ/বছর |
-|---|---|---|
-| সম্পূর্ণ কনটেক্সট পেস্ট করা | 19.5M+ | অসম্ভব (উইন্ডো অতিক্রম করে) |
-| LLM-সংক্ষেপিত | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + লোকাল এমবেডিং | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> এমবেডিং মডেল: `all-MiniLM-L6-v2` (লোকাল, ফ্রি, API কী ছাড়াই)। সম্পূর্ণ রিপোর্ট: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md)। প্রতিযোগীদের সাথে তুলনা: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), যা agentmemory-কে mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo-র সাথে তুলনা করে।
-
-**লোকালি রিপ্রোডিউস করুন:** [`eval/README.md`](../eval/README.md), LongMemEval `_s` (পাবলিক 500-প্রশ্ন) + `coding-agent-life-v1` (ইন-হাউস 15-সেশন করপাস)-এর জন্য একটি অ্যাডাপ্টার-পাগেবল হার্নেস। grep / vector / agentmemory অ্যাডাপ্টারগুলো পাশাপাশি স্কোর করে, NDJSON আউটপুট দেয়, প্রকাশিত স্কোরকার্ড [`docs/benchmarks/`](../docs/benchmarks/)-এ জমা হয়।
-
-**[codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), এবং [Graphify](https://github.com/safishamsi/graphify)-এর সাথে ভালোভাবে জুটি বাঁধে।** কোড-গ্রাফ ইনডেক্সিং, মাল্টি-এজেন্ট বিল্ড পাইপলাইন, এবং docs / PDF / ছবি / ভিডিও জুড়ে বিস্তৃত নলেজ গ্রাফ। agentmemory কাজটা মনে রাখে; এই তিনটি প্রজেক্ট কনটেক্সট লেয়ারের বাকি অংশ আলোকিত করে। রেসিপি + প্রশ্ন-রাউটিং টেবিল: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md)।
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="প্রতিযোগীদের সাথে তুলনা" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>বিল্ট-ইন (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>ধরন</strong></td>
-<td>মেমরি ইঞ্জিন + MCP সার্ভার</td>
-<td>মেমরি লেয়ার API</td>
-<td>সম্পূর্ণ এজেন্ট রানটাইম</td>
-<td>পার্সোনাল AI</td>
-<td>মেমরি API + অ্যাপ</td>
-<td>টিম মেমরি হাব (LLM প্রক্সি)</td>
-<td>ভেক্টর মেমরি (OSS)</td>
-<td>মেমরি ইঞ্জিন (Oracle DB)</td>
-<td>মেমরি সিস্টেম</td>
-<td>স্ট্যাটিক ফাইল</td>
-</tr>
-<tr>
-<td><strong>রিট্রিভাল R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>স্ব-ঘোষিত</td>
-<td>PersonaMem 76% (স্ব-ঘোষিত)</td>
-<td>~96.6% (স্ব-ঘোষিত)</td>
-<td>94.4% (স্ব-ঘোষিত)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>অটো-ক্যাপচার</strong></td>
-<td>12 হুক (ম্যানুয়াল পরিশ্রম শূন্য)</td>
-<td>ম্যানুয়াল <code>add()</code> কল</td>
-<td>এজেন্ট সেলফ-এডিট</td>
-<td>ম্যানুয়াল</td>
-<td>API-সাইড এক্সট্রাকশন</td>
-<td>প্রক্সি ইন্টারসেপশন (base-URL সোয়াপ)</td>
-<td>ম্যানুয়াল</td>
-<td>API এক্সট্রাকশন</td>
-<td>ম্যানুয়াল</td>
-<td>ম্যানুয়াল এডিটিং</td>
-</tr>
-<tr>
-<td><strong>সার্চ</strong></td>
-<td>BM25 + Vector + Graph (RRF ফিউশন)</td>
-<td>Vector + Graph</td>
-<td>Vector (আর্কাইভাল)</td>
-<td>সিম্যান্টিক</td>
-<td>Vector + RAG</td>
-<td>4 ধরনের অ্যাসেট (Chat / Skill / Wiki / CodeGraph)</td>
-<td>কেবল Vector</td>
-<td>Vector + সিম্যান্টিক</td>
-<td>ডিকে-ওয়েটেড</td>
-<td>সব কনটেক্সটে লোড করে</td>
-</tr>
-<tr>
-<td><strong>মাল্টি-এজেন্ট</strong></td>
-<td>MCP + REST + lease + signal</td>
-<td>API (কোনো কো-অর্ডিনেশন নেই)</td>
-<td>কেবল Letta রানটাইমের মধ্যে</td>
-<td>না</td>
-<td>না</td>
-<td>টিম রোল + শেয়ার্ড অ্যাসেট</td>
-<td>না</td>
-<td>কেবল স্কোপড</td>
-<td>মাল্টি-এজেন্ট শেয়ার্ড</td>
-<td>প্রতি-এজেন্ট ফাইল</td>
-</tr>
-<tr>
-<td><strong>ফ্রেমওয়ার্ক লক-ইন</strong></td>
-<td>নেই (যেকোনো MCP ক্লায়েন্ট)</td>
-<td>নেই</td>
-<td>বেশি (Letta-ই ব্যবহার করতে হবে)</td>
-<td>স্ট্যান্ডঅ্যালোন</td>
-<td>নেই</td>
-<td>প্রক্সি প্রতিটি মডেল কলের সামনে থাকে</td>
-<td>নেই</td>
-<td>Oracle Database</td>
-<td>নেই</td>
-<td>প্রতি-এজেন্ট ফরম্যাট</td>
-</tr>
-<tr>
-<td><strong>এক্সটার্নাল ডিপেন্ডেন্সি</strong></td>
-<td>আলাদা ডেটাবেস সেবা দরকার নেই</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + ভেক্টর DB</td>
-<td>একাধিক</td>
-<td>ম্যানেজড ক্লাউড</td>
-<td>Docker স্ট্যাক (Core + Hub + Proxy)</td>
-<td>ভেক্টর স্টোর</td>
-<td>Oracle AI Database</td>
-<td>নেই</td>
-<td>নেই</td>
-</tr>
-<tr>
-<td><strong>মেমরি লাইফসাইকেল</strong></td>
-<td>4-টায়ার কনসোলিডেশন + ডিকে + অটো-ফরগেট</td>
-<td>প্যাসিভ এক্সট্রাকশন</td>
-<td>এজেন্ট-ম্যানেজড</td>
-<td>ম্যানুয়াল</td>
-<td>অটো-ফরগেট</td>
-<td>ম্যানুয়াল রিভিউ; অটো-রাউটিং চলছে</td>
-<td>নেই</td>
-<td>বলা নেই</td>
-<td>ডিকে + কনসোলিডেশন</td>
-<td>ম্যানুয়াল প্রুনিং</td>
-</tr>
-<tr>
-<td><strong>টোকেন এফিশিয়েন্সি</strong></td>
-<td>~1,900 টোকেন/সেশন ($10/বছর)</td>
-<td>ইন্টিগ্রেশন অনুযায়ী ভিন্ন</td>
-<td>কোর মেমরি কনটেক্সটে থাকে</td>
-<td>ভিন্ন হয়</td>
-<td>ক্লাউড প্রাইসিং</td>
-<td>বলা নেই</td>
-<td>কোনো টোকেন বাজেট নেই</td>
-<td>LLM-ব্যাকড (ভিন্ন হয়)</td>
-<td>ভিন্ন হয়</td>
-<td>240 অবজারভেশনে 22K+ টোকেন</td>
-</tr>
-<tr>
-<td><strong>রিয়েল-টাইম ভিউয়ার</strong></td>
-<td>হ্যাঁ (পোর্ট 3113)</td>
-<td>ক্লাউড ড্যাশবোর্ড</td>
-<td>ক্লাউড ড্যাশবোর্ড</td>
-<td>ওয়েব UI</td>
-<td>ক্লাউড ড্যাশবোর্ড</td>
-<td>হাব ওয়েব UI</td>
-<td>না</td>
-<td>না</td>
-<td>না</td>
-<td>না</td>
-</tr>
-<tr>
-<td><strong>সেলফ-হোস্টেড</strong></td>
-<td>হ্যাঁ (ডিফল্ট)</td>
-<td>অপশনাল</td>
-<td>অপশনাল</td>
-<td>হ্যাঁ</td>
-<td>না (কেবল ক্লাউড)</td>
-<td>হ্যাঁ (Docker)</td>
-<td>হ্যাঁ</td>
-<td>হ্যাঁ (Oracle DB)</td>
-<td>হ্যাঁ</td>
-<td>হ্যাঁ</td>
-</tr>
-</table>
-
-<sub>বেঞ্চমার্ক নোট: শুধু agentmemory-র R@5 আমাদের নিজের মাপা ফল (LongMemEval-S, <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> থেকে রিপ্রোডিউসযোগ্য)। mem0 এবং Letta-র সংখ্যাগুলো তাদের প্রকাশিত LoCoMo নম্বর (একটি ভিন্ন ডেটাসেট); MemPalace, supermemory, TencentDB (PersonaMem), এবং oracleagentmemory-র সংখ্যাগুলো ভেন্ডরের স্ব-ঘোষিত দাবি যা আমরা স্বতন্ত্রভাবে রিপ্রোডিউস করিনি (oracleagentmemory-র রানটি একটি Oracle AI Database-এর বিরুদ্ধে GPT-5.5 ব্যবহার করেছিল)। পাশাপাশি দেখানো হয়েছে কেবল মোটামুটি ধারণা দেওয়ার জন্য, অভিন্ন ডেটার উপর মুখোমুখি তুলনা নয়। স্টার সংখ্যা আনুমানিক এবং সময়ের সাথে বদলায়।</sub>
-
-**নতুন প্রতিযোগী** যাদের জানা দরকার, [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md)-এ গভীরভাবে তুলনা করা হয়েছে:
-
-| সিস্টেম | ⭐ | অ্যাঙ্গেল |
-|--------|---|-------|
-| Zep / Graphiti | 30K | টেম্পোরাল নলেজ গ্রাফ; সবচেয়ে শক্তিশালী প্রকাশিত টেম্পোরাল-কোয়েরি ফল (LongMemEval 63.8%), কিন্তু গ্রাফ অ্যাসিনক্রোনাসভাবে তৈরি হয় তাই নতুন তথ্য দেরিতে আসতে পারে |
-| Cognee | 30K | ডকুমেন্ট-থেকে-নলেজ-গ্রাফ ইনজেশন, কেবল Python, সেশন ক্যাপচারের বদলে স্ট্রাকচারড এনটিটি এক্সট্র্যাকশনের জন্য তৈরি |
+পণ্য ও স্থাপনার বিকল্পের হালনাগাদ তুলনা এবং উৎস:
-AgentMemory সেশন ক্যাপচার, অনুসন্ধানযোগ্য প্রজেক্ট ইতিহাস ও লোকাল ভিউয়ার একসঙ্গে দেয়; প্রথমবার থেকেই কী ছাড়া ব্যবহার করা যায়।
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
মেমরি সার্ভার শুরু করুন: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add`-এর মাধ্যমে নেটিভ স্কিল (50+ এজেন্ট)
+#### `npx skills add`-এর মাধ্যমে নেটিভ স্কিল
agentmemory Claude-Code-স্টাইলের `<dir>/SKILL.md` ফরম্যাটে 17টি স্কিল নিয়ে আসে: 9টি ইনভোকেবল অ্যাকশন স্কিল (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) এবং 8টি রেফারেন্স স্কিল যা এজেন্ট প্রয়োজনমতো লোড করে (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`)। রেফারেন্স স্কিলগুলো সোর্স থেকে জেনারেট করা ডেটা টেবিল বহন করে, তাই এগুলো কখনো ডিফ্ট করে না। vercel-labs-এর [`skills`](https://npmjs.com/package/skills) CLI 50+ এজেন্টে (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, এবং আরও) কলিং এজেন্টের নেটিভ স্কিল ডিরেক্টরিতে এগুলো অটো-ইনস্টল করে:
@@ -837,7 +629,7 @@ Hermes, OpenClaw, Pi ও Oh My Pi কানেক্টর ডিফল্টভ
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` শেয়ার্ড কাস্টমাইজেশন ডিরেক্টরিতে MCP এবং ক্যাপচার হুক ইনস্টল করে। দেখুন [Antigravity সেটআপ ও সীমাবদ্ধতা](../docs/plugins/antigravity.md)। |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` বর্তমান IDE ভার্সনের মতো একই MCP এবং হুক কনফিগারেশন ব্যবহার করে। বিদ্যমান ইনস্টলেশনগুলো `--force` দিয়ে রিফ্রেশ করা উচিত; দেখুন [আপগ্রেড নোটস](../docs/plugins/antigravity.md)। |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` ইউজার-লেভেল কনফিগ লেখে। ওয়ার্কস্পেস ওভাররাইড আপনার কোডের পাশে `.kiro/settings/mcp.json`-এ যায়। |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` স্ট্যান্ডার্ড `mcpServers` ব্লক লেখে। Warp `.claude/skills/` থেকেও স্কিল অটো-ডিসকভার করে; Claude Code প্লাগইন ইনস্টল হলে agentmemory-র 8টি স্কিল (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) Warp-এর স্ল্যাশ-কমান্ড প্যালেটে নেটিভভাবে দেখা যায়। |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` স্ট্যান্ডার্ড `mcpServers` ব্লক লেখে। Warp `.claude/skills/` থেকেও স্কিল অটো-ডিসকভার করে; Claude Code প্লাগইন ইনস্টল হলে agentmemory-র 8টি স্কিল (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) Warp-এর স্ল্যাশ-কমান্ড প্যালেটে নেটিভভাবে দেখা যায়। |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` স্ট্যান্ডার্ড `mcpServers` ব্লক লেখে। VS Code এক্সটেনশন ইউজাররা: Cline Settings → MCP Servers → Edit JSON-এর মাধ্যমে একই ব্লক পেস্ট করুন। |
| **Continue.dev** | `~/.continue/config.yaml` (পছন্দনীয়) বা `config.json` (লেগ্যাসি) | `agentmemory connect continue` কনফিগারেশন না থাকলে `config.yaml` তৈরি করে অথবা বর্তমান `config.json` আপডেট করে। বিদ্যমান YAML-এর comment ও anchor বাঁচাতে সেটি না বদলে `mcpServers:`-এর নিচে বসানোর block দেখায়। Continue-এর `mcpServers` একটি array। |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` `context_servers`-এর নিচে লেখে (Zed-র কি, `mcpServers` নয়)। রিমোট MCP সার্ভার এর বদলে `{"url": "..."}` দিয়েও ওয়্যার করা যায়। |
@@ -933,7 +725,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| রক্ষণাবেক্ষণ | হাতে সাজানো নির্দেশনা | ইনডেক্স করা সেশন ইতিহাস ও মেমরি |
| সার্চ | সবকিছু কনটেক্সটে লোড করে | BM25 + vector + graph (কেবল top-K) |
-| টোকেন খরচ | 240 অবজারভেশনে 22K+ | ~1,900 টোকেন (92% কম) |
| ক্রস-এজেন্ট | প্রতি-এজেন্ট ফাইল | MCP + REST (যেকোনো এজেন্ট) |
| কো-অর্ডিনেশন | নেই | lease, signal, action, routine |
| অবজারভেবিলিটি | ম্যানুয়ালি ফাইল পড়া | :3113-এ রিয়েল-টাইম ভিউয়ার |
@@ -946,7 +737,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Reciprocal Rank Fusion (RRF, k=60) দিয়ে ফিউজড এবং স
ভেক্টর ইনডেক্স পপুলেটেড থাকলে, `mem::search` (`memory_recall`-এর পেছনে) হাইব্রিড BM25 + vector র্যাংকার ব্যবহার করে। এমবেডিং ছাড়া এটি BM25 ব্যবহার করে। `smart-search` গ্রাফ ডেটা থাকলে অতিরিক্তভাবে স্ট্রাকচারাল গ্রাফ ম্যাচ ফিউজ করতে পারে, কিলেস মোডেও। লেসন রিকল প্রতি কোয়েরিতে সম্পূর্ণ করপাস স্ক্যান করার বদলে একটি ডেডিকেটেড ইন-মেমরি BM25 ইনডেক্সে চলে। রিপ্লেসড মেমরি ভার্সন প্রতিটি রিকল পাথ থেকে বাদ পড়ে; ভার্সন চেইন তাদের ইতিহাস রাখে।
-ভেক্টর ক্র্যাশ বা ফোর্স-কিল সহ্য করতে পারে। ভেক্টর ইনডেক্স সর্বোচ্চ প্রতি `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 মিনিট) অন্তর বাকেটে সেভ হয়। মাঝখানে যুক্ত বা বাদ দেওয়া প্রতিটি ভেক্টরও state store-এ একটি ছোট পেন্ডিং লগে তাৎক্ষণিকভাবে লেখা হয়, এবং পরের স্টার্ট এমবেডিং প্রোভাইডারকে কল না করেই সেটা রিপ্লে করে। প্রতিটি সফল সেভ লগটি খালি করে দেয়। রিপ্লের পরেও যে ডকুমেন্টগুলোর ভেক্টর নেই সেগুলো ব্যাকগ্রাউন্ডে `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500)-এর ব্যাচে পুনরায়-এমবেড করা হয় যতক্ষণ না কোনোটা বাকি থাকে, এবং থেমে যাওয়া ব্যাকফিল পরের স্টার্টে চালু থাকে। `/agentmemory/status` এবং ভিউয়ার পেন্ডিং লগের সাইজ ও ব্যাকফিল স্টেট দেখায়। কিলেস ইনস্টল কিছুই লেখে না।
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 আউট অফ দ্য বক্সে গ্রিক, সিরিলিক, হিব্রু, আরবি, এবং অ্যাকসেন্টেড ল্যাটিন টোকেনাইজ করে। চাইনিজ / জাপানিজ / কোরিয়ান মেমরির জন্য, CJK রানগুলো ওয়ার্ড-লেভেল টোকেনে ভাগ করতে অপশনাল সেগমেন্টার ইনস্টল করুন (`npm install @node-rs/jieba tiny-segmenter`); এগুলো ছাড়া, agentmemory সফট-ফল করে সম্পূর্ণ-রান টোকেনাইজেশনে এবং stderr-এ একবার একটি হিন্ট প্রিন্ট করে।
@@ -1047,9 +838,9 @@ EMBEDDING_PROVIDER=local
| প্রোভাইডার | মডেল | খরচ | নোট |
|---|---|---|---|
-| **Local (সুপারিশকৃত অপ্ট-ইন)** | `all-MiniLM-L6-v2` | ফ্রি | প্রথম মডেল ডাউনলোডের পর অন-ডিভাইস, শুধু-BM25-র তুলনায় +8pp রিকল |
+| **Local (সুপারিশকৃত অপ্ট-ইন)** | `all-MiniLM-L6-v2` | ফ্রি | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | ফ্রি টায়ার | 100+ ভাষা, 768/1536/3072 dims (MRL), 2048-টোকেন ইনপুট। `text-embedding-004`-কে প্রতিস্থাপন করে ([deprecated, শাটডাউন Jan 14, 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | সর্বোচ্চ কোয়ালিটি |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | পেইড | কোডের জন্য অপ্টিমাইজড |
| Cohere | `embed-english-v3.0` | ফ্রি ট্রায়াল | জেনারেল পারপাস |
| OpenRouter | কোনো মডেল | ভিন্ন হয় | মাল্টি-মডেল প্রক্সি |
diff --git a/READMEs/README.cs-CZ.md b/READMEs/README.cs-CZ.md
index 2442c41f..fde107da 100644
--- a/READMEs/README.cs-CZ.md
+++ b/READMEs/README.cs-CZ.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="o 92% méně tokenů" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP nástrojů" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automatických hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 externích databází" height="38" /></picture>
@@ -333,223 +331,17 @@ Nejnovější poznámky k vydání: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarky" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Přesnost vyhledávání
-
-**coding-agent-life-v1** (vlastní korpus, reprodukovatelný v sandboxu)
-
-| Adaptér | P@5 | R@5 | Úspěšnost top-5 | latence p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% úspěšnost top-5 na **matematickém stropu P@5** pro tento korpus (0.240, viz scorecard). Hybrid najde každou zlatou session; grep mine 1 ze 2 zlatých u vícesessionového temporálního dotazu. Zisk je v **recall + temporalitě**, ne v agregátní přesnosti. Tento benchmark je malý a s málo zlatými případy; větší LongMemEval-S níže rozlišuje lépe. Úplný rozpad podle typu + oprava: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 otázek)
+Aktuální výsledky vyhledávání, definice metrik a postup reprodukce:
-| Systém | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| záložní režim pouze BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Úspora tokenů
-
-| Přístup | Tokenů/rok | Náklady/rok |
-|---|---|---|
-| Vložení celého kontextu | 19.5M+ | Nemožné (překračuje okno) |
-| Shrnuto LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokální embeddingy | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding model: `all-MiniLM-L6-v2` (lokální, zdarma, bez API klíče). Úplné reporty: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Srovnání s konkurencí: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) porovnává agentmemory s mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reprodukce lokálně:** [`eval/README.md`](../eval/README.md), harness s zapojitelnými adaptéry pro LongMemEval `_s` (veřejných 500 otázek) + `coding-agent-life-v1` (vlastní korpus 15 sessions). Adaptéry grep / vektor / agentmemory bodují vedle sebe, výstup NDJSON, publikované scorecardy jsou v [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Sedí dobře s [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) a [Graphify](https://github.com/safishamsi/graphify).** Indexace grafu kódu, vícagentní build pipeline a širší znalostní grafy napříč dokumenty / PDF / obrázky / videi. agentmemory si pamatuje práci; tyto tři projekty rozsvítí zbytek kontextové vrstvy. Recepty + tabulka směrování podle otázky: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Srovnání s konkurencí" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Vestavěné (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Typ</strong></td>
-<td>Paměťový engine + server MCP</td>
-<td>API paměťové vrstvy</td>
-<td>Plný runtime agenta</td>
-<td>Osobní AI</td>
-<td>API paměti + aplikace</td>
-<td>Týmový hub paměti (LLM proxy)</td>
-<td>Vektorová paměť (OSS)</td>
-<td>Paměťový engine (Oracle DB)</td>
-<td>Systém paměti</td>
-<td>Statický soubor</td>
-</tr>
-<tr>
-<td><strong>Vyhledávání R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Údaj výrobce</td>
-<td>PersonaMem 76% (údaj výrobce)</td>
-<td>~96.6% (údaj výrobce)</td>
-<td>94.4% (údaj výrobce)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Automatické zachycení</strong></td>
-<td>12 hooks (bez manuálního úsilí)</td>
-<td>Manuální volání <code>add()</code></td>
-<td>Agent si upravuje sám</td>
-<td>Manuální</td>
-<td>Extrakce na straně API</td>
-<td>Zachytávání přes proxy (výměna base URL)</td>
-<td>Manuální</td>
-<td>Extrakce přes API</td>
-<td>Manuální</td>
-<td>Manuální úpravy</td>
-</tr>
-<tr>
-<td><strong>Vyhledávání</strong></td>
-<td>BM25 + vektor + graf (fúze RRF)</td>
-<td>Vektor + graf</td>
-<td>Vektor (archivní)</td>
-<td>Sémantické</td>
-<td>Vektor + RAG</td>
-<td>4 typy assetů (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Pouze vektor</td>
-<td>Vektor + sémantika</td>
-<td>Váženo rozpadem (decay)</td>
-<td>Načte vše do kontextu</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signály</td>
-<td>API (bez koordinace)</td>
-<td>Pouze uvnitř runtime Letta</td>
-<td>Ne</td>
-<td>Ne</td>
-<td>Týmové role + sdílené assety</td>
-<td>Ne</td>
-<td>Pouze v rámci rozsahu</td>
-<td>Sdíleno mezi agenty</td>
-<td>Soubory po agentech</td>
-</tr>
-<tr>
-<td><strong>Vázanost na framework</strong></td>
-<td>Žádná (jakýkoli klient MCP)</td>
-<td>Žádná</td>
-<td>Vysoká (musíte používat Letta)</td>
-<td>Samostatné</td>
-<td>Žádná</td>
-<td>Proxy stojí před každým voláním modelu</td>
-<td>Žádná</td>
-<td>Oracle Database</td>
-<td>Žádná</td>
-<td>Formát po agentech</td>
-</tr>
-<tr>
-<td><strong>Externí závislosti</strong></td>
-<td>Není potřeba samostatná databázová služba</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektorová DB</td>
-<td>Více</td>
-<td>Správovaný cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vektorové úložiště</td>
-<td>Oracle AI Database</td>
-<td>Žádné</td>
-<td>Žádné</td>
-</tr>
-<tr>
-<td><strong>Životní cyklus paměti</strong></td>
-<td>4úrovňová konsolidace + rozpad + automatické zapomínání</td>
-<td>Pasivní extrakce</td>
-<td>Správa agentem</td>
-<td>Manuální</td>
-<td>Automatické zapomínání</td>
-<td>Manuální revize; automatické směrování se připravuje</td>
-<td>Žádný</td>
-<td>Neuvedeno</td>
-<td>Rozpad + konsolidace</td>
-<td>Manuální prořezávání</td>
-</tr>
-<tr>
-<td><strong>Efektivita tokenů</strong></td>
-<td>~1,900 tokenů/session ($10/rok)</td>
-<td>Liší se podle integrace</td>
-<td>Klíčová paměť v kontextu</td>
-<td>Liší se</td>
-<td>Cloudové ceny</td>
-<td>Neuvedeno</td>
-<td>Žádný rozpočet tokenů</td>
-<td>Založeno na LLM (liší se)</td>
-<td>Liší se</td>
-<td>22K+ tokenů při 240 obs.</td>
-</tr>
-<tr>
-<td><strong>Viewer v reálném čase</strong></td>
-<td>Ano (port 3113)</td>
-<td>Cloudový dashboard</td>
-<td>Cloudový dashboard</td>
-<td>Webové UI</td>
-<td>Cloudový dashboard</td>
-<td>Webové UI hubu</td>
-<td>Ne</td>
-<td>Ne</td>
-<td>Ne</td>
-<td>Ne</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Ano (výchozí)</td>
-<td>Volitelné</td>
-<td>Volitelné</td>
-<td>Ano</td>
-<td>Ne (pouze cloud)</td>
-<td>Ano (Docker)</td>
-<td>Ano</td>
-<td>Ano (Oracle DB)</td>
-<td>Ano</td>
-<td>Ano</td>
-</tr>
-</table>
-
-<sub>Poznámka k benchmarku: pouze R@5 pro agentmemory je náš vlastní naměřený výsledek (LongMemEval-S, reprodukovatelný z <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Čísla pro mem0 a Letta jsou jejich publikované hodnoty LoCoMo (jiný dataset); čísla pro MemPalace, supermemory, TencentDB (PersonaMem) a oracleagentmemory jsou tvrzení výrobců, která jsme nezávisle nereprodukovali (běh oracleagentmemory použil GPT-5.5 proti Oracle AI Database). Uvedeno vedle sebe jen pro orientační srovnání, nejde o přímé porovnání na identických datech. Počty hvězd jsou orientační a v čase se mění.</sub>
-
-**Novější konkurenti**, o kterých má smysl vědět, podrobně srovnáni v [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Systém | ⭐ | Přístup |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporální znalostní graf; nejsilnější publikované výsledky pro temporální dotazy (LongMemEval 63.8%), ale graf se buduje asynchronně, takže čerstvá fakta mohou zpožďovat |
-| Cognee | 30K | Načítání dokumentů do znalostního grafu, pouze Python, navrženo pro strukturovanou extrakci entit, nikoli pro zachytávání sessions |
+Aktuální srovnání produktů a možností nasazení se zdroji:
-AgentMemory spojuje zachycení sessions, prohledávatelnou historii projektu a lokální viewer; bez klíče funguje od prvního spuštění.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
Spusťte server paměti: `npx -y @agentmemory/agentmemory@latest`
-#### Nativní skills přes `npx skills add` (50+ agentů)
+#### Nativní skills přes `npx skills add`
agentmemory dodává 17 skills ve formátu `<dir>/SKILL.md` ve stylu Claude Code: 9 vyvolatelných akčních skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) a 8 referenčních skills, které agent načítá podle potřeby (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Referenční skills obsahují datové tabulky generované ze zdroje, takže se nikdy neliší. CLI [`skills`](https://npmjs.com/package/skills) od vercel-labs je automaticky instaluje do nativního adresáře skills volajícího agenta napříč 50+ agenty (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf a dalšími):
@@ -844,7 +636,7 @@ Položka agentmemory je **stejný blok serveru MCP** na všech hostitelích, kte
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` nainstaluje MCP a hooky pro zachycení do sdíleného adresáře přizpůsobení. Viz [nastavení a limity Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` používá stejnou konfiguraci MCP a hooků jako aktuální verze IDE. Stávající instalace by měly provést refresh pomocí `--force`; viz [poznámky k upgradu](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` zapíše konfiguraci na úrovni uživatele. Přepisy na úrovni workspace patří do `.kiro/settings/mcp.json` vedle vašeho kódu. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` zapíše standardní blok `mcpServers`. Warp také automaticky vyhledává skills z `.claude/skills/`; po instalaci pluginu Claude Code se 8 skills agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) objeví nativně v paletě slash příkazů Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` zapíše standardní blok `mcpServers`. Warp také automaticky vyhledává skills z `.claude/skills/`; po instalaci pluginu Claude Code se 9 skills agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) objeví nativně v paletě slash příkazů Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` zapíše standardní blok `mcpServers`. Uživatelé rozšíření VS Code: vložte stejný blok přes Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferované) nebo `config.json` (starší) | `agentmemory connect continue` vytvoří `config.yaml`, pokud konfigurace chybí, nebo upraví existující `config.json`. U existujícího YAML vypíše blok k vložení pod `mcpServers:`, aby zachoval komentáře a kotvy. Continue používá pole, nikoli objekt. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` zapíše pod `context_servers` (klíč Zed, NE `mcpServers`). Vzdálené servery MCP lze zapojit místo toho přes `{"url": "..."}`. |
@@ -927,7 +719,6 @@ Každý AI kódovací agent se dodává s vestavěnou pamětí: Claude Code má
|---|---|---|
| Údržba | Ručně spravované instrukce | Indexovaná historie sessions a vzpomínky |
| Vyhledávání | načte vše do kontextu | BM25 + vektor + graf (pouze top-K) |
-| Náklady na tokeny | 22K+ při 240 pozorováních | ~1,900 tokenů (o 92% méně) |
| Napříč agenty | soubory po agentech | MCP + REST (jakýkoli agent) |
| Koordinace | žádná | leases, signály, akce, routiny |
| Observabilita | manuální čtení souborů | viewer v reálném čase na :3113 |
@@ -940,7 +731,7 @@ Každý AI kódovací agent se dodává s vestavěnou pamětí: Claude Code má
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ Sloučeno pomocí Reciprocal Rank Fusion (RRF, k=60) a diverzifikováno po sessi
Když je vektorový index naplněný, `mem::search` (za `memory_recall`) používá hybridní řadič BM25 + vektor. Bez embeddingů používá BM25. `smart-search` může navíc sloučit strukturální shody z grafu, pokud grafová data existují, včetně režimu bez klíče. Vyhledávání lessons běží na vyhrazeném in-memory indexu BM25 místo skenování celého korpusu při každém dotazu. Nahrazené verze paměti jsou vyloučeny z každé cesty vyhledávání; řetězec verzí uchovává jejich historii.
-Vektory přežijí pád nebo násilné vypnutí. Vektorový index se ukládá v dávkách nejvýše jednou za `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minut). Každý vektor přidaný nebo odstraněný mezitím se navíc okamžitě zapíše do malého čekajícího logu ve state store a při dalším startu se přehraje bez volání poskytovatele embeddingů. Každé úspěšné uložení log vyprázdní. Dokumenty, které po přehrání stále nemají vektor, se na pozadí znovu embedují v dávkách po `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), dokud žádné nezbydou, a zastavený backfill pokračuje při dalším startu. `/agentmemory/status` a viewer zobrazují velikost čekajícího logu a stav backfillu. Instalace bez klíče nic nezapisují.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 standardně tokenizuje řečtinu, cyrilici, hebrejštinu, arabštinu a akcentovanou latinku. Pro paměti v čínštině / japonštině / korejštině nainstalujte volitelné segmentery (`npm install @node-rs/jieba tiny-segmenter`), aby se běhy CJK rozdělily na tokeny na úrovni slov; bez nich agentmemory měkce spadne na tokenizaci celého běhu a jednou vypíše nápovědu na stderr.
@@ -1041,9 +832,9 @@ Běžná instalace npm obsahuje volitelný runtime `@huggingface/transformers`.
| Poskytovatel | Model | Náklady | Poznámky |
|---|---|---|---|
-| **Lokální (doporučeno jako opt-in)** | `all-MiniLM-L6-v2` | zdarma | na zařízení po prvním stažení modelu, +8pp recall oproti samotnému BM25 |
+| **Lokální (doporučeno jako opt-in)** | `all-MiniLM-L6-v2` | zdarma | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | bezplatný tier | 100+ jazyků, rozměry 768/1536/3072 (MRL), vstup 2048 tokenů. Nahrazuje `text-embedding-004` ([deprecated, ukončení 14. 1. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | nejvyšší kvalita |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | placené | optimalizováno pro kód |
| Cohere | `embed-english-v3.0` | bezplatná zkouška | univerzální použití |
| OpenRouter | jakýkoli model | liší se | proxy pro více modelů |
diff --git a/READMEs/README.da-DK.md b/READMEs/README.da-DK.md
index 13dced17..5eac7f2f 100644
--- a/READMEs/README.da-DK.md
+++ b/READMEs/README.da-DK.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% færre tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP-værktøjer" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automatiske hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 eksterne databaser" height="38" /></picture>
@@ -331,223 +329,17 @@ Seneste udgivelsesnoter: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval-nøjagtighed
-
-**coding-agent-life-v1** (internt korpus, reproducerbart i sandbox)
-
-| Adapter | P@5 | R@5 | Top-5-hitrate | p50-latens |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% top-5-hitrate ved **P@5-matematikkens loft** for dette korpus (0.240, se scorecard). Hybrid finder hver gylden session; grep misser 1 af 2 gyldne i den multi-session-temporale forespørgsel. Forbedringen er **recall + temporal**, ikke samlet precision. Dette benchmark er lille og har få gyldne eksempler; det større LongMemEval-S nedenfor differentierer bedre. Fuld opdeling pr. type + korrektionsnote: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 spørgsmål)
+Aktuelle søgeresultater, definitioner af målinger og trin til gentagelse:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Kun BM25-fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tokenbesparelser
-
-| Metode | Tokens/år | Omkostning/år |
-|---|---|---|
-| Indsæt fuld kontekst | 19.5M+ | Umuligt (overstiger konteksten) |
-| LLM-opsummeret | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokale embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding-model: `all-MiniLM-L6-v2` (lokal, gratis, ingen API-nøgle). Fulde rapporter: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Konkurrentsammenligning: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), der dækker agentmemory vs. mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproducér lokalt:** [`eval/README.md`](../eval/README.md), en adapter-plugbar harness til LongMemEval `_s` (offentligt 500-spørgsmåls-sæt) + `coding-agent-life-v1` (internt korpus med 15 sessioner). Grep-, vektor- og agentmemory-adaptere scorer side om side, NDJSON-output, publicerede scorecards lander i [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Fungerer godt sammen med [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) og [Graphify](https://github.com/safishamsi/graphify).** Code-graph-indeksering, multi-agent build-pipelines og bredere videngrafer tværs af dokumenter/PDF'er/billeder/videoer. agentmemory husker arbejdet; disse tre projekter oplyser resten af kontekstlaget. Opskrifter + spørgsmålsrouting-tabel: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs. konkurrenter" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Indbygget (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Type</strong></td>
-<td>Hukommelsesmotor + MCP-server</td>
-<td>Hukommelseslag-API</td>
-<td>Fuld agent-runtime</td>
-<td>Personlig AI</td>
-<td>Hukommelses-API + app</td>
-<td>Team-hukommelseshub (LLM-proxy)</td>
-<td>Vektorhukommelse (OSS)</td>
-<td>Hukommelsesmotor (Oracle DB)</td>
-<td>Hukommelsessystem</td>
-<td>Statisk fil</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Selvrapporteret</td>
-<td>PersonaMem 76% (selvrapporteret)</td>
-<td>~96.6% (selvrapporteret)</td>
-<td>94.4% (selvrapporteret)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Automatisk indsamling</strong></td>
-<td>12 hooks (ingen manuel indsats)</td>
-<td>Manuelle <code>add()</code>-kald</td>
-<td>Agenten redigerer selv</td>
-<td>Manuel</td>
-<td>Udtrækning på API-siden</td>
-<td>Proxy-aflytning (ombytning af base-URL)</td>
-<td>Manuel</td>
-<td>API-udtrækning</td>
-<td>Manuel</td>
-<td>Manuel redigering</td>
-</tr>
-<tr>
-<td><strong>Søgning</strong></td>
-<td>BM25 + vektor + graf (RRF-fusion)</td>
-<td>Vektor + graf</td>
-<td>Vektor (arkiv)</td>
-<td>Semantisk</td>
-<td>Vektor + RAG</td>
-<td>4 aktivtyper (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Kun vektor</td>
-<td>Vektor + semantisk</td>
-<td>Forfald-vægtet</td>
-<td>Indlæser alt i konteksten</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signaler</td>
-<td>API (ingen koordinering)</td>
-<td>Kun inden for Letta-runtime</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Teamroller + delte aktiver</td>
-<td>Nej</td>
-<td>Kun scoped</td>
-<td>Delt mellem flere agenter</td>
-<td>Filer pr. agent</td>
-</tr>
-<tr>
-<td><strong>Framework-lock-in</strong></td>
-<td>Ingen (enhver MCP-klient)</td>
-<td>Ingen</td>
-<td>Høj (skal bruge Letta)</td>
-<td>Selvstændig</td>
-<td>Ingen</td>
-<td>Proxy foran hvert modelkald</td>
-<td>Ingen</td>
-<td>Oracle Database</td>
-<td>Ingen</td>
-<td>Format pr. agent</td>
-</tr>
-<tr>
-<td><strong>Eksterne afhængigheder</strong></td>
-<td>Ingen separat databasetjeneste nødvendig</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektordatabase</td>
-<td>Flere</td>
-<td>Administreret cloud</td>
-<td>Docker-stack (Core + Hub + Proxy)</td>
-<td>Vektorlager</td>
-<td>Oracle AI Database</td>
-<td>Ingen</td>
-<td>Ingen</td>
-</tr>
-<tr>
-<td><strong>Hukommelseslivscyklus</strong></td>
-<td>4-lags konsolidering + forfald + auto-forglemmelse</td>
-<td>Passiv udtrækning</td>
-<td>Agent-styret</td>
-<td>Manuel</td>
-<td>Auto-forglemmelse</td>
-<td>Manuel gennemgang; auto-routing under udvikling</td>
-<td>Ingen</td>
-<td>Ikke angivet</td>
-<td>Forfald + konsolidering</td>
-<td>Manuel beskæring</td>
-</tr>
-<tr>
-<td><strong>Tokeneffektivitet</strong></td>
-<td>~1,900 tokens/session ($10/år)</td>
-<td>Varierer efter integration</td>
-<td>Core-hukommelse i konteksten</td>
-<td>Varierer</td>
-<td>Cloud-prissætning</td>
-<td>Ikke angivet</td>
-<td>Intet tokenbudget</td>
-<td>LLM-baseret (varierer)</td>
-<td>Varierer</td>
-<td>22K+ tokens ved 240 observationer</td>
-</tr>
-<tr>
-<td><strong>Realtids-viewer</strong></td>
-<td>Ja (port 3113)</td>
-<td>Cloud-dashboard</td>
-<td>Cloud-dashboard</td>
-<td>Web-UI</td>
-<td>Cloud-dashboard</td>
-<td>Hub web-UI</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Nej</td>
-</tr>
-<tr>
-<td><strong>Selvhostet</strong></td>
-<td>Ja (standard)</td>
-<td>Valgfrit</td>
-<td>Valgfrit</td>
-<td>Ja</td>
-<td>Nej (kun cloud)</td>
-<td>Ja (Docker)</td>
-<td>Ja</td>
-<td>Ja (Oracle DB)</td>
-<td>Ja</td>
-<td>Ja</td>
-</tr>
-</table>
-
-<sub>Benchmark-note: kun agentmemorys R@5 er vores eget målte resultat (LongMemEval-S, reproducerbart fra <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Tallene for mem0 og Letta er deres publicerede LoCoMo-tal (et andet datasæt); tallene for MemPalace, supermemory, TencentDB (PersonaMem) og oracleagentmemory er leverandørens selvrapporterede påstande, som vi ikke selv har reproduceret (oracleagentmemorys kørsel brugte GPT-5.5 mod en Oracle AI Database). Vist side om side kun som et groft overblik, ikke en direkte sammenligning på identiske data. Stjerneantal er cirka og ændrer sig over tid.</sub>
-
-**Nyere aktører**, som er værd at kende, sammenlignet i detaljer i [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | Vinkel |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal videngraf; stærkeste publicerede resultater for temporale forespørgsler (LongMemEval 63.8%), men grafen bygges asynkront, så nye fakta kan halte |
-| Cognee | 30K | Dokument-til-videngraf-indlæsning, kun Python, bygget til struktureret entitetsudtrækning snarere end sessionsindsamling |
+Opdateret sammenligning af produkter og driftsmuligheder med kilder:
-AgentMemory kombinerer registrering fra kodningsagenter, søgbar projekthistorik og en lokal viewer. Du kan arbejde uden nøgle fra første start.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Fuld guide: [`integrations/hermes/`](../integrations/hermes/)
Start hukommelsesserveren: `npx -y @agentmemory/agentmemory@latest`
-#### Native skills via `npx skills add` (50+ agenter)
+#### Native skills via `npx skills add`
agentmemory leveres med 17 skills i Claude-Code-stilens `<dir>/SKILL.md`-format: 9 kaldbare action-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) og 8 referenceskills, som agenten indlæser efter behov (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Referenceskillene indeholder datatabeller genereret fra kildekoden, så de aldrig driver ud af kurs. [`skills`](https://npmjs.com/package/skills)-CLI'en fra vercel-labs installerer dem automatisk i den kaldende agents native skill-mappe på tværs af 50+ agenter (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf og flere):
@@ -836,7 +628,7 @@ agentmemory-indgangen er den **samme MCP-serverblok** på alle hosts, der bruger
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installerer MCP og capture-hooks i den delte tilpasningsmappe. Se [Antigravity-opsætning og begrænsninger](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` bruger samme MCP- og hook-konfiguration som de nuværende IDE-versioner. Eksisterende installationer bør opdateres med `--force`; se [opgraderingsnoterne](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` skriver konfigurationen på brugerniveau. Workspace-overrides placeres i `.kiro/settings/mcp.json` ved siden af din kode. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver standard-`mcpServers`-blokken. Warp opdager også automatisk skills fra `.claude/skills/`; når Claude Code-pluginet er installeret, vises de 8 agentmemory-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) nativt i Warps slash-kommando-palette. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver standard-`mcpServers`-blokken. Warp opdager også automatisk skills fra `.claude/skills/`; når Claude Code-pluginet er installeret, vises de 9 agentmemory-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) nativt i Warps slash-kommando-palette. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` skriver standard-`mcpServers`-blokken. Brugere af VS Code-extensionen: indsæt samme blok via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (foretrukket) eller `config.json` (legacy) | `agentmemory connect continue` opretter en manglende `config.yaml` eller ændrer eksisterende `config.json`. Hvis YAML-filen findes, vises blokken til `mcpServers:`, så kommentarer og ankre bevares. Continue bruger et array, ikke et objekt. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` skriver under `context_servers` (Zeds nøgle, IKKE `mcpServers`). Eksterne MCP-servere kan i stedet forbindes via `{"url": "..."}`. |
@@ -933,7 +725,6 @@ Hver AI-kodeagent leveres med indbygget hukommelse: Claude Code har `MEMORY.md`,
|---|---|---|
| Vedligeholdelse | Manuelt vedligeholdte instruktioner | Indekseret sessionshistorik og hukommelse |
| Søgning | Indlæser alt i konteksten | BM25 + vektor + graf (kun top-K) |
-| Tokenomkostning | 22K+ ved 240 observationer | ~1,900 tokens (92% mindre) |
| Tværs af agenter | Filer pr. agent | MCP + REST (enhver agent) |
| Koordinering | Ingen | Leases, signaler, actions, routines |
| Observerbarhed | Læs filer manuelt | Realtids-viewer på :3113 |
@@ -946,7 +737,7 @@ Hver AI-kodeagent leveres med indbygget hukommelse: Claude Code har `MEMORY.md`,
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Fusioneret med Reciprocal Rank Fusion (RRF, k=60) og session-diversificeret (mak
Når et vektorindeks er udfyldt, bruger `mem::search` (bag `memory_recall`) den hybride BM25 + vektor-ranker. Uden embeddings bruger den BM25. `smart-search` kan desuden fusionere strukturelle graf-match, når der findes graf-data, også i nøgleløs tilstand. Lesson-recall køres på et dedikeret in-memory BM25-indeks i stedet for at scanne hele korpuset pr. forespørgsel. Overskrevne hukommelsesversioner udelukkes fra alle recall-veje; versionskæden bevarer deres historik.
-Vektorer overlever et crash eller force-kill. Vektorindekset gemmes i buckets højst hver `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutter). Hver vektor, der tilføjes eller fjernes imellemtiden, skrives også straks til en lille ventende log i state-store'en, og den næste opstart afspiller den uden at kalde embedding-udbyderen. Hver gennemført save tømmer loggen. Dokumenter, der stadig ikke har en vektor efter afspilningen, genindlejres i baggrunden i batches af `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), indtil ingen er tilbage, og en backfill, der stoppes, fortsætter ved næste opstart. `/agentmemory/status` og viewer viser størrelsen af den ventende log og backfill-tilstanden. Nøgleløse installationer skriver ingenting.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokeniserer græsk, kyrillisk, hebraisk, arabisk og accentueret latin ud af boksen. For kinesiske/japanske/koreanske hukommelser skal du installere de valgfrie segmenterere (`npm install @node-rs/jieba tiny-segmenter`) for at splitte CJK-forløb i ord-niveau-tokens; uden dem falder agentmemory blidt tilbage til hel-forløb-tokenisering og printer et engangshint på stderr.
@@ -1047,9 +838,9 @@ Den normale npm-installation inkluderer den valgfrie `@huggingface/transformers`
| Udbyder | Model | Pris | Noter |
|---|---|---|---|
-| **Lokal (anbefalet tilvalg)** | `all-MiniLM-L6-v2` | Gratis | På enheden efter første modeldownload, +8pp recall over kun BM25 |
+| **Lokal (anbefalet tilvalg)** | `all-MiniLM-L6-v2` | Gratis | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Gratis niveau | 100+ sprog, 768/1536/3072 dimensioner (MRL), 2048-token input. Afløser `text-embedding-004` ([udfaset, lukket 14. januar 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Højeste kvalitet |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Betalt | Optimeret til kode |
| Cohere | `embed-english-v3.0` | Gratis prøveperiode | Generelt anvendelig |
| OpenRouter | Enhver model | Varierer | Multi-model-proxy |
diff --git a/READMEs/README.de-DE.md b/READMEs/README.de-DE.md
index 16413332..311e1459 100644
--- a/READMEs/README.de-DE.md
+++ b/READMEs/README.de-DE.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95,2 % Retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92 % weniger Tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP-Tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 Auto-Hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 externe DBs" height="38" /></picture>
@@ -329,223 +327,17 @@ Neueste Release-Hinweise: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval-Genauigkeit
-
-**coding-agent-life-v1** (interner Korpus, Sandbox-reproduzierbar)
-
-| Adapter | P@5 | R@5 | Top-5-Trefferquote | p50-Latenz |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-Baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100 % Top-5-Trefferquote an der **mathematischen P@5-Obergrenze** für diesen Korpus (0.240, siehe Scorecard). Hybrid findet jede Gold-Session; grep verfehlt 1 von 2 Gold-Sessions bei der Multi-Session-Temporalanfrage. Der Gewinn ist **Recall + Temporal**, nicht aggregierte Präzision. Dieser Benchmark ist klein und Gold-arm; das größere LongMemEval-S unten differenziert besser. Vollständige Aufschlüsselung pro Typ + Korrekturhinweis: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 Fragen)
+Aktuelle Retrieval-Ergebnisse, Metrikdefinitionen und Schritte zur Reproduktion:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Nur-BM25-Fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token-Einsparungen
-
-| Ansatz | Tokens/Jahr | Kosten/Jahr |
-|---|---|---|
-| Vollständigen Kontext einfügen | 19.5M+ | Unmöglich (überschreitet das Fenster) |
-| LLM-zusammengefasst | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokale Embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding-Modell: `all-MiniLM-L6-v2` (lokal, kostenlos, kein API-Schlüssel). Vollständige Berichte: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Konkurrenzvergleich: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), der agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo abdeckt.
-
-**Lokal reproduzieren:** [`eval/README.md`](../eval/README.md), ein Adapter-pluggable Harness für LongMemEval `_s` (öffentlich, 500 Fragen) + `coding-agent-life-v1` (interner 15-Session-Korpus). Adapter für Grep / Vector / agentmemory werden direkt verglichen, NDJSON-Ausgabe, veröffentlichte Scorecards landen in [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Funktioniert kombiniert mit [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) und [Graphify](https://github.com/safishamsi/graphify).** Code-Graph-Indizierung, mehragentige Build-Pipelines und breitere Knowledge Graphs über Docs / PDFs / Bilder / Videos. agentmemory merkt sich die Arbeit; diese drei Projekte beleuchten den Rest der Kontextschicht. Rezepte + Frage-Routing-Tabelle: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Vergleich mit der Konkurrenz" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Eingebaut (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Typ</strong></td>
-<td>Memory-Engine + MCP-Server</td>
-<td>Memory-Layer-API</td>
-<td>Komplette Agenten-Runtime</td>
-<td>Persönliche KI</td>
-<td>Memory-API + App</td>
-<td>Team-Memory-Hub (LLM-Proxy)</td>
-<td>Vector-Memory (OSS)</td>
-<td>Memory-Engine (Oracle DB)</td>
-<td>Memory-System</td>
-<td>Statische Datei</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/V</td>
-<td>Selbst berichtet</td>
-<td>PersonaMem 76% (selbst berichtet)</td>
-<td>~96.6% (selbst berichtet)</td>
-<td>94.4% (selbst berichtet)</td>
-<td>N/V</td>
-<td>N/V (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-Erfassung</strong></td>
-<td>12 Hooks (null manueller Aufwand)</td>
-<td>Manuelle <code>add()</code>-Aufrufe</td>
-<td>Agent bearbeitet sich selbst</td>
-<td>Manuell</td>
-<td>API-seitige Extraktion</td>
-<td>Proxy-Interception (Base-URL-Tausch)</td>
-<td>Manuell</td>
-<td>API-Extraktion</td>
-<td>Manuell</td>
-<td>Manuelle Bearbeitung</td>
-</tr>
-<tr>
-<td><strong>Suche</strong></td>
-<td>BM25 + Vector + Graph (RRF-Fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (Archival)</td>
-<td>Semantisch</td>
-<td>Vector + RAG</td>
-<td>4 Asset-Typen (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Nur Vector</td>
-<td>Vector + semantisch</td>
-<td>Decay-gewichtet</td>
-<td>Lädt alles in den Kontext</td>
-</tr>
-<tr>
-<td><strong>Multi-Agent</strong></td>
-<td>MCP + REST + Leases + Signals</td>
-<td>API (keine Koordination)</td>
-<td>Nur innerhalb der Letta-Runtime</td>
-<td>Nein</td>
-<td>Nein</td>
-<td>Team-Rollen + geteilte Assets</td>
-<td>Nein</td>
-<td>Nur Scoped</td>
-<td>Multi-Agent geteilt</td>
-<td>Dateien pro Agent</td>
-</tr>
-<tr>
-<td><strong>Framework-Lock-in</strong></td>
-<td>Keiner (jeder MCP-Client)</td>
-<td>Keiner</td>
-<td>Hoch (Letta erforderlich)</td>
-<td>Standalone</td>
-<td>Keiner</td>
-<td>Proxy sitzt vor jedem Modellaufruf</td>
-<td>Keiner</td>
-<td>Oracle Database</td>
-<td>Keiner</td>
-<td>Format pro Agent</td>
-</tr>
-<tr>
-<td><strong>Externe Abhängigkeiten</strong></td>
-<td>Kein separater Datenbankdienst erforderlich</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + Vector-DB</td>
-<td>Mehrere</td>
-<td>Managed Cloud</td>
-<td>Docker-Stack (Core + Hub + Proxy)</td>
-<td>Vector-Store</td>
-<td>Oracle AI Database</td>
-<td>Keine</td>
-<td>Keine</td>
-</tr>
-<tr>
-<td><strong>Memory-Lifecycle</strong></td>
-<td>4-stufige Konsolidierung + Decay + Auto-Forget</td>
-<td>Passive Extraktion</td>
-<td>Vom Agenten verwaltet</td>
-<td>Manuell</td>
-<td>Auto-Forget</td>
-<td>Manuelles Review; Auto-Routing in Arbeit</td>
-<td>Keiner</td>
-<td>Nicht angegeben</td>
-<td>Decay + Konsolidierung</td>
-<td>Manuelles Pruning</td>
-</tr>
-<tr>
-<td><strong>Token-Effizienz</strong></td>
-<td>~1,900 Tokens/Session ($10/Jahr)</td>
-<td>Je nach Integration unterschiedlich</td>
-<td>Core Memory im Kontext</td>
-<td>Variiert</td>
-<td>Cloud-Preise</td>
-<td>Nicht angegeben</td>
-<td>Kein Token-Budget</td>
-<td>LLM-gestützt (variiert)</td>
-<td>Variiert</td>
-<td>22K+ Tokens bei 240 Beobachtungen</td>
-</tr>
-<tr>
-<td><strong>Echtzeit-Viewer</strong></td>
-<td>Ja (Port 3113)</td>
-<td>Cloud-Dashboard</td>
-<td>Cloud-Dashboard</td>
-<td>Web-UI</td>
-<td>Cloud-Dashboard</td>
-<td>Hub-Web-UI</td>
-<td>Nein</td>
-<td>Nein</td>
-<td>Nein</td>
-<td>Nein</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Ja (Standard)</td>
-<td>Optional</td>
-<td>Optional</td>
-<td>Ja</td>
-<td>Nein (nur Cloud)</td>
-<td>Ja (Docker)</td>
-<td>Ja</td>
-<td>Ja (Oracle DB)</td>
-<td>Ja</td>
-<td>Ja</td>
-</tr>
-</table>
-
-<sub>Benchmark-Hinweis: Nur agentmemorys R@5 ist unser eigenes gemessenes Ergebnis (LongMemEval-S, reproduzierbar aus <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Die Zahlen von mem0 und Letta sind deren veröffentlichte LoCoMo-Werte (ein anderer Datensatz); die Zahlen von MemPalace, supermemory, TencentDB (PersonaMem) und oracleagentmemory sind selbst berichtete Herstellerangaben, die wir nicht unabhängig reproduziert haben (der Lauf von oracleagentmemory verwendete GPT-5.5 gegen eine Oracle AI Database). Nebeneinander nur zur groben Einordnung gezeigt, kein direkter Vergleich auf identischen Daten. Star-Zahlen sind ungefähr und driften über die Zeit.</sub>
-
-**Neuere Einsteiger**, die man kennen sollte, ausführlich verglichen in [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | Ausrichtung |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporaler Knowledge Graph; stärkste veröffentlichte Temporal-Query-Ergebnisse (LongMemEval 63.8%), aber der Graph wird asynchron aufgebaut, sodass frische Fakten hinterherhinken können |
-| Cognee | 30K | Dokument-zu-Knowledge-Graph-Ingestion, nur Python, gebaut für strukturierte Entitäten-Extraktion statt Session-Erfassung |
+Aktueller Vergleich von Produkten und Bereitstellungsoptionen mit Quellen:
-AgentMemory verbindet die Erfassung durch Coding-Agenten mit durchsuchbarer Projekthistorie und einem lokalen Viewer. Der Betrieb ohne API-Schlüssel ist vom ersten Start an möglich.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -777,7 +569,7 @@ Vollständiger Leitfaden: [`integrations/hermes/`](../integrations/hermes/)
Starten Sie den Memory-Server: `npx -y @agentmemory/agentmemory@latest`
-#### Native Skills via `npx skills add` (50+ Agenten)
+#### Native Skills via `npx skills add`
agentmemory liefert 17 Skills im Claude-Code-artigen `<dir>/SKILL.md`-Format: 9 aufrufbare Action-Skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) und 8 Referenz-Skills, die der Agent bei Bedarf lädt (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Die Referenz-Skills tragen aus dem Quellcode generierte Datentabellen, sodass sie nie driften. Die [`skills`](https://npmjs.com/package/skills)-CLI von vercel-labs installiert sie automatisch in das native Skill-Verzeichnis des aufrufenden Agenten, über 50+ Agenten hinweg (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf und mehr):
@@ -834,7 +626,7 @@ Der agentmemory-Eintrag ist der **gleiche MCP-Server-Block** für jeden Host, de
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installiert MCP und Capture-Hooks im gemeinsam genutzten Customization-Verzeichnis. Siehe [Antigravity-Setup und -Grenzen](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` verwendet dieselbe MCP- und Hook-Konfiguration wie aktuelle IDE-Versionen. Bestehende Installationen sollten mit `--force` aktualisieren; siehe die [Upgrade-Hinweise](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` schreibt die Konfig auf Benutzerebene. Workspace-Overrides liegen in `.kiro/settings/mcp.json` neben Ihrem Code. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` schreibt den standardmäßigen `mcpServers`-Block. Warp entdeckt außerdem Skills aus `.claude/skills/` automatisch; sobald das Claude-Code-Plugin installiert ist, erscheinen die 8 agentmemory-Skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) nativ in Warps Slash-Command-Palette. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` schreibt den standardmäßigen `mcpServers`-Block. Warp entdeckt außerdem Skills aus `.claude/skills/` automatisch; sobald das Claude-Code-Plugin installiert ist, erscheinen die 9 agentmemory-Skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) nativ in Warps Slash-Command-Palette. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` schreibt den standardmäßigen `mcpServers`-Block. Nutzer der VS-Code-Extension: Fügen Sie denselben Block über Cline Settings → MCP Servers → Edit JSON ein. |
| **Continue.dev** | `~/.continue/config.yaml` (bevorzugt) oder `config.json` (Legacy) | `agentmemory connect continue` erstellt eine fehlende `config.yaml` oder aktualisiert eine vorhandene `config.json`. Bei bestehender YAML-Datei zeigt der Adapter den Block für `mcpServers:` zum Einfügen an; Kommentare und Anker bleiben erhalten. Continue verwendet dort ein Array, kein Objekt. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` schreibt unter `context_servers` (Zeds Schlüssel, NICHT `mcpServers`). Remote-MCP-Server können stattdessen via `{"url": "..."}` verdrahtet werden. |
@@ -931,7 +723,6 @@ Jeder KI-Coding-Agent kommt mit eingebautem Memory: Claude Code hat `MEMORY.md`,
|---|---|---|
| Pflege | Manuell gepflegte Anweisungen | Indizierte Session-Historie und Erinnerungen |
| Suche | Lädt alles in den Kontext | BM25 + Vector + Graph (nur Top-K) |
-| Token-Kosten | 22K+ bei 240 Beobachtungen | ~1,900 Tokens (92 % weniger) |
| Agentenübergreifend | Dateien pro Agent | MCP + REST (jeder Agent) |
| Koordination | Keine | Leases, Signale, Actions, Routinen |
| Observability | Dateien manuell lesen | Echtzeit-Viewer auf :3113 |
@@ -944,7 +735,7 @@ Jeder KI-Coding-Agent kommt mit eingebautem Memory: Claude Code hat `MEMORY.md`,
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1029,7 +820,7 @@ Verschmolzen mit Reciprocal Rank Fusion (RRF, k=60) und session-diversifiziert (
Wenn ein Vector-Index befüllt ist, verwendet `mem::search` (hinter `memory_recall`) den hybriden BM25-+-Vector-Ranker. Ohne Embeddings verwendet es BM25. `smart-search` kann zusätzlich strukturelle Graph-Treffer einbeziehen, wenn Graph-Daten existieren, auch im schlüssellosen Modus. Lesson-Recall läuft auf einem dedizierten In-Memory-BM25-Index, statt bei jeder Anfrage den ganzen Korpus zu scannen. Überholte Memory-Versionen sind von jedem Recall-Pfad ausgeschlossen; die Versionskette bewahrt ihre Historie.
-Vectors überleben einen Absturz oder Force-Kill. Der Vector-Index wird in Buckets gespeichert, höchstens alle `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 Minuten). Jeder dazwischen hinzugefügte oder entfernte Vector wird außerdem sofort in ein kleines Pending-Log im State-Store geschrieben, und der nächste Start spielt es ab, ohne den Embedding-Provider aufzurufen. Jedes erfolgreiche Speichern leert das Log. Dokumente, die nach dem Replay noch keinen Vector haben, werden im Hintergrund in Batches von `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) neu eingebettet, bis keine mehr übrig sind, und ein gestopptes Backfill wird beim nächsten Start fortgesetzt. `/agentmemory/status` und der Viewer zeigen die Größe des Pending-Logs und den Backfill-Status. Schlüssellose Installationen schreiben nichts.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenisiert Griechisch, Kyrillisch, Hebräisch, Arabisch und akzentuiertes Latein standardmäßig. Für Erinnerungen in Chinesisch / Japanisch / Koreanisch installieren Sie die optionalen Segmentierer (`npm install @node-rs/jieba tiny-segmenter`), um CJK-Folgen in Worttokens aufzuteilen; ohne sie fällt agentmemory weich auf eine Tokenisierung als gesamte Folge zurück und gibt einmalig einen Hinweis auf stderr aus.
@@ -1045,9 +836,9 @@ Die normale npm-Installation enthält die optionale `@huggingface/transformers`-
| Provider | Modell | Kosten | Hinweise |
|---|---|---|---|
-| **Lokal (empfohlenes Opt-in)** | `all-MiniLM-L6-v2` | Kostenlos | On-device nach dem ersten Modell-Download, +8pp Recall gegenüber Nur-BM25 |
+| **Lokal (empfohlenes Opt-in)** | `all-MiniLM-L6-v2` | Kostenlos | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free Tier | 100+ Sprachen, 768/1536/3072 Dims (MRL), 2048-Token-Eingabe. Ersetzt `text-embedding-004` ([deprecated, Abschaltung 14. Jan. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Höchste Qualität |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Kostenpflichtig | Auf Code optimiert |
| Cohere | `embed-english-v3.0` | Testzugang | Allzweck |
| OpenRouter | Beliebiges Modell | Variabel | Multi-Modell-Proxy |
diff --git a/READMEs/README.el-GR.md b/READMEs/README.el-GR.md
index fa2254a8..a021d9cf 100644
--- a/READMEs/README.el-GR.md
+++ b/READMEs/README.el-GR.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% ανάκληση R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% λιγότερα tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 εργαλεία MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 αυτόματα hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 εξωτερικές βάσεις δεδομένων" height="38" /></picture>
@@ -333,223 +331,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Ακρίβεια Ανάκλησης
-
-**coding-agent-life-v1** (εσωτερικό corpus, αναπαραγώγιμο σε sandbox)
-
-| Προσαρμογέας | P@5 | R@5 | Ποσοστό επιτυχίας Top-5 | Καθυστέρηση p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep βασικής αναφοράς | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Ποσοστό επιτυχίας Top-5 100% στο **ανώτατο μαθηματικό όριο P@5** για αυτό το corpus (0.240, δείτε το scorecard). Το hybrid ανακτά κάθε gold συνεδρία· το grep χάνει 1 από τα 2 gold στο ερώτημα πολλαπλών συνεδριών με χρονική διάσταση. Η βελτίωση είναι σε **ανάκληση + χρονικότητα**, όχι σε συνολική ακρίβεια. Αυτό το benchmark είναι μικρό και με λίγα gold δεδομένα· το μεγαλύτερο LongMemEval-S παρακάτω διαφοροποιεί καλύτερα. Πλήρης ανάλυση ανά τύπο + σημείωση διόρθωσης: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 ερωτήσεις)
+Τρέχοντα αποτελέσματα ανάκτησης, ορισμοί μετρικών και βήματα αναπαραγωγής:
-| Σύστημα | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Μόνο-BM25 (fallback) | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Εξοικονόμηση Tokens
-
-| Προσέγγιση | Tokens/έτος | Κόστος/έτος |
-|---|---|---|
-| Επικόλληση πλήρους context | 19.5M+ | Αδύνατο (υπερβαίνει το παράθυρο) |
-| Σύνοψη από LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + τοπικά embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Μοντέλο embedding: `all-MiniLM-L6-v2` (τοπικό, δωρεάν, χωρίς κλειδί API). Πλήρεις αναφορές: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Σύγκριση με ανταγωνιστές: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), που καλύπτει το agentmemory έναντι των mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Αναπαραγωγή τοπικά:** [`eval/README.md`](../eval/README.md), ένα harness με συνδέσιμους adapters για το LongMemEval `_s` (δημόσιο, 500 ερωτήσεις) + `coding-agent-life-v1` (εσωτερικό corpus 15 συνεδριών). Οι adapters grep / vector / agentmemory βαθμολογούνται παράλληλα, με έξοδο NDJSON, και τα δημοσιευμένα scorecards καταλήγουν στο [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Συνεργάζεται με τα [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), και [Graphify](https://github.com/safishamsi/graphify).** Ευρετηρίαση code-graph, pipelines κατασκευής πολλών agents, και ευρύτεροι γράφοι γνώσης σε docs / PDF / εικόνες / βίντεο. Το agentmemory θυμάται τη δουλειά· αυτά τα τρία projects αναδεικνύουν την υπόλοιπη context layer. Συνταγές + πίνακας δρομολόγησης ερωτημάτων: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Έναντι Ανταγωνιστών" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Ενσωματωμένο (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Τύπος</strong></td>
-<td>Μηχανή μνήμης + διακομιστής MCP</td>
-<td>API επιπέδου μνήμης</td>
-<td>Πλήρες runtime agent</td>
-<td>Προσωπικό AI</td>
-<td>API μνήμης + εφαρμογή</td>
-<td>Κόμβος μνήμης ομάδας (proxy LLM)</td>
-<td>Vector μνήμη (OSS)</td>
-<td>Μηχανή μνήμης (Oracle DB)</td>
-<td>Σύστημα μνήμης</td>
-<td>Στατικό αρχείο</td>
-</tr>
-<tr>
-<td><strong>Ανάκληση R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Αυτοαναφερόμενο</td>
-<td>PersonaMem 76% (αυτοαναφερόμενο)</td>
-<td>~96.6% (αυτοαναφερόμενο)</td>
-<td>94.4% (αυτοαναφερόμενο)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Αυτόματη καταγραφή</strong></td>
-<td>12 hooks (καμία χειροκίνητη προσπάθεια)</td>
-<td>Χειροκίνητες κλήσεις <code>add()</code></td>
-<td>Αυτο-επεξεργασίες από τον agent</td>
-<td>Χειροκίνητο</td>
-<td>Εξαγωγή από την πλευρά του API</td>
-<td>Παρεμβολή proxy (αλλαγή base-URL)</td>
-<td>Χειροκίνητο</td>
-<td>Εξαγωγή API</td>
-<td>Χειροκίνητο</td>
-<td>Χειροκίνητη επεξεργασία</td>
-</tr>
-<tr>
-<td><strong>Αναζήτηση</strong></td>
-<td>BM25 + Vector + Graph (συγχώνευση RRF)</td>
-<td>Vector + Graph</td>
-<td>Vector (αρχειοθέτηση)</td>
-<td>Σημασιολογική</td>
-<td>Vector + RAG</td>
-<td>4 τύποι assets (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Μόνο Vector</td>
-<td>Vector + σημασιολογική</td>
-<td>Σταθμισμένη βάσει decay</td>
-<td>Φορτώνει τα πάντα στο context</td>
-</tr>
-<tr>
-<td><strong>Πολλαπλών agents</strong></td>
-<td>MCP + REST + leases + σήματα</td>
-<td>API (χωρίς συντονισμό)</td>
-<td>Μόνο εντός του runtime Letta</td>
-<td>Όχι</td>
-<td>Όχι</td>
-<td>Ρόλοι ομάδας + κοινά assets</td>
-<td>Όχι</td>
-<td>Μόνο εντός scope</td>
-<td>Κοινό μεταξύ πολλών agents</td>
-<td>Αρχεία ανά agent</td>
-</tr>
-<tr>
-<td><strong>Εξάρτηση από framework</strong></td>
-<td>Καμία (οποιοδήποτε MCP client)</td>
-<td>Καμία</td>
-<td>Υψηλή (απαιτεί χρήση Letta)</td>
-<td>Αυτόνομο</td>
-<td>Καμία</td>
-<td>Το proxy μεσολαβεί σε κάθε κλήση μοντέλου</td>
-<td>Καμία</td>
-<td>Oracle Database</td>
-<td>Καμία</td>
-<td>Μορφή ανά agent</td>
-</tr>
-<tr>
-<td><strong>Εξωτερικές εξαρτήσεις</strong></td>
-<td>Δεν απαιτείται ξεχωριστή υπηρεσία βάσης δεδομένων</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>Πολλαπλές</td>
-<td>Διαχειριζόμενο cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Αποθήκη vector</td>
-<td>Oracle AI Database</td>
-<td>Καμία</td>
-<td>Καμία</td>
-</tr>
-<tr>
-<td><strong>Κύκλος ζωής μνήμης</strong></td>
-<td>Ενοποίηση 4 επιπέδων + decay + αυτόματη διαγραφή</td>
-<td>Παθητική εξαγωγή</td>
-<td>Διαχειρίζεται από τον agent</td>
-<td>Χειροκίνητο</td>
-<td>Αυτόματη διαγραφή</td>
-<td>Χειροκίνητη επισκόπηση· αυτόματη δρομολόγηση σε εξέλιξη</td>
-<td>Καμία</td>
-<td>Δεν αναφέρεται</td>
-<td>Decay + ενοποίηση</td>
-<td>Χειροκίνητο κλάδεμα</td>
-</tr>
-<tr>
-<td><strong>Απόδοση tokens</strong></td>
-<td>~1,900 tokens/συνεδρία ($10/έτος)</td>
-<td>Διαφέρει ανά ολοκλήρωση</td>
-<td>Βασική μνήμη εντός context</td>
-<td>Διαφέρει</td>
-<td>Τιμολόγηση cloud</td>
-<td>Δεν αναφέρεται</td>
-<td>Χωρίς προϋπολογισμό tokens</td>
-<td>Βασισμένο σε LLM (διαφέρει)</td>
-<td>Διαφέρει</td>
-<td>22K+ tokens σε 240 παρατηρήσεις</td>
-</tr>
-<tr>
-<td><strong>Viewer σε πραγματικό χρόνο</strong></td>
-<td>Ναι (θύρα 3113)</td>
-<td>Dashboard cloud</td>
-<td>Dashboard cloud</td>
-<td>Web UI</td>
-<td>Dashboard cloud</td>
-<td>Hub web UI</td>
-<td>Όχι</td>
-<td>Όχι</td>
-<td>Όχι</td>
-<td>Όχι</td>
-</tr>
-<tr>
-<td><strong>Αυτοφιλοξενούμενο (self-hosted)</strong></td>
-<td>Ναι (προεπιλογή)</td>
-<td>Προαιρετικό</td>
-<td>Προαιρετικό</td>
-<td>Ναι</td>
-<td>Όχι (μόνο cloud)</td>
-<td>Ναι (Docker)</td>
-<td>Ναι</td>
-<td>Ναι (Oracle DB)</td>
-<td>Ναι</td>
-<td>Ναι</td>
-</tr>
-</table>
-
-<sub>Σημείωση benchmark: μόνο το R@5 του agentmemory είναι δικό μας μετρημένο αποτέλεσμα (LongMemEval-S, αναπαραγώγιμο από το <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Τα στοιχεία των mem0 και Letta είναι οι δημοσιευμένες τιμές τους στο LoCoMo (διαφορετικό dataset)· τα στοιχεία των MemPalace, supermemory, TencentDB (PersonaMem), και oracleagentmemory είναι αυτοαναφερόμενοι ισχυρισμοί προμηθευτών που δεν έχουμε αναπαράγει ανεξάρτητα (η εκτέλεση του oracleagentmemory χρησιμοποίησε GPT-5.5 έναντι μιας Oracle AI Database). Παρουσιάζονται δίπλα-δίπλα μόνο για κατά προσέγγιση σύγκριση, όχι ως απευθείας αντιπαράθεση σε ταυτόσημα δεδομένα. Οι μετρήσεις αστεριών είναι κατά προσέγγιση και μεταβάλλονται με τον χρόνο.</sub>
-
-**Νεότεροι παίκτες** που αξίζει να γνωρίζετε, συγκρίνονται αναλυτικά στο [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Σύστημα | ⭐ | Προσέγγιση |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Χρονικός γράφος γνώσης· τα ισχυρότερα δημοσιευμένα αποτελέσματα σε χρονικά ερωτήματα (LongMemEval 63.8%), αλλά ο γράφος χτίζεται ασύγχρονα, οπότε τα νέα δεδομένα μπορεί να καθυστερούν |
-| Cognee | 30K | Εισαγωγή εγγράφων σε γράφο γνώσης, μόνο σε Python, σχεδιασμένο για δομημένη εξαγωγή οντοτήτων παρά για καταγραφή συνεδριών |
+Ενημερωμένη σύγκριση προϊόντων και επιλογών εγκατάστασης με πηγές:
-Το AgentMemory συνδυάζει καταγραφή συνεδριών, αναζητήσιμο ιστορικό έργου και τοπικό viewer, με λειτουργία χωρίς κλειδί από την πρώτη εκκίνηση.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
Ξεκινήστε τον διακομιστή μνήμης: `npx -y @agentmemory/agentmemory@latest`
-#### Εγγενή skills μέσω `npx skills add` (50+ agents)
+#### Εγγενή skills μέσω `npx skills add`
Το agentmemory διανέμει 17 skills στη μορφή `<dir>/SKILL.md` τύπου Claude Code: 9 επικλητά skills ενέργειας (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) και 8 skills αναφοράς που ο agent φορτώνει όταν τα χρειάζεται (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Τα skills αναφοράς περιέχουν πίνακες δεδομένων που δημιουργούνται από την πηγή, οπότε δεν αποκλίνουν ποτέ. Το CLI [`skills`](https://npmjs.com/package/skills) από την vercel-labs τα εγκαθιστά αυτόματα στον εγγενή κατάλογο skills του agent που το καλεί, σε 50+ agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, και άλλα):
@@ -844,7 +636,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | Το `agentmemory connect antigravity --with-hooks` εγκαθιστά το MCP και τα hooks καταγραφής στον κοινόχρηστο κατάλογο προσαρμογής. Δείτε [ρύθμιση και περιορισμοί Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | Το `agentmemory connect antigravity-cli --with-hooks` χρησιμοποιεί την ίδια ρύθμιση MCP και hooks με τις τρέχουσες εκδόσεις IDE. Οι υπάρχουσες εγκαταστάσεις πρέπει να ανανεωθούν με `--force`· δείτε τις [σημειώσεις αναβάθμισης](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | Το `agentmemory connect kiro` γράφει τις ρυθμίσεις σε επίπεδο χρήστη. Οι παρακάμψεις workspace μπαίνουν στο `.kiro/settings/mcp.json` δίπλα στον κώδικά σας. |
-| **Warp** | `~/.warp/.mcp.json` | Το `agentmemory connect warp` γράφει το τυπικό block `mcpServers`. Το Warp επίσης ανακαλύπτει αυτόματα skills από το `.claude/skills/`· μόλις εγκατασταθεί το plugin του Claude Code, τα 8 skills του agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) εμφανίζονται εγγενώς στην παλέτα εντολών slash του Warp. |
+| **Warp** | `~/.warp/.mcp.json` | Το `agentmemory connect warp` γράφει το τυπικό block `mcpServers`. Το Warp επίσης ανακαλύπτει αυτόματα skills από το `.claude/skills/`· μόλις εγκατασταθεί το plugin του Claude Code, τα 9 skills του agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) εμφανίζονται εγγενώς στην παλέτα εντολών slash του Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | Το `agentmemory connect cline` γράφει το τυπικό block `mcpServers`. Χρήστες της extension VS Code: επικολλήστε το ίδιο block μέσω Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (προτιμώμενο) ή `config.json` (παλαιό) | Το `agentmemory connect continue` δημιουργεί `config.yaml` αν λείπουν οι ρυθμίσεις ή αλλάζει υπάρχον `config.json`. Για υπάρχον YAML εμφανίζει το block προς προσθήκη στο `mcpServers:`, διατηρώντας σχόλια και anchors. Το Continue χρησιμοποιεί πίνακα, όχι αντικείμενο. |
| **Zed** | `~/.config/zed/settings.json` | Το `agentmemory connect zed` γράφει κάτω από το `context_servers` (το κλειδί του Zed, ΟΧΙ `mcpServers`). Απομακρυσμένοι διακομιστές MCP μπορούν να συνδεθούν εναλλακτικά μέσω `{"url": "..."}`. |
@@ -927,7 +719,6 @@ npx -y @agentmemory/mcp
|---|---|---|
| Συντήρηση | Χειροκίνητα διαχειριζόμενες οδηγίες | Ευρετηριασμένο ιστορικό συνεδριών και μνήμες |
| Αναζήτηση | Φορτώνει τα πάντα στο context | BM25 + vector + graph (μόνο top-K) |
-| Κόστος tokens | 22K+ σε 240 παρατηρήσεις | ~1,900 tokens (92% λιγότερο) |
| Μεταξύ agents | Αρχεία ανά agent | MCP + REST (οποιοσδήποτε agent) |
| Συντονισμός | Κανένας | Leases, signals, actions, routines |
| Παρατηρησιμότητα | Χειροκίνητη ανάγνωση αρχείων | Viewer σε πραγματικό χρόνο στη :3113 |
@@ -940,7 +731,7 @@ npx -y @agentmemory/mcp
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ SessionStart hook fires
Όταν ένα ευρετήριο vector είναι γεμάτο, το `mem::search` (πίσω από το `memory_recall`) χρησιμοποιεί τον υβριδικό ranker BM25 + vector. Χωρίς embeddings χρησιμοποιεί BM25. Το `smart-search` μπορεί επιπλέον να συγχωνεύσει δομικές αντιστοιχίσεις γράφου όταν υπάρχουν δεδομένα γράφου, ακόμα και σε λειτουργία χωρίς κλειδί. Η ανάκληση μαθημάτων (lesson recall) τρέχει σε ένα αποκλειστικό in-memory ευρετήριο BM25 αντί να σκανάρει όλο το corpus σε κάθε ερώτημα. Οι αντικατεστημένες εκδόσεις μνήμης αποκλείονται από κάθε διαδρομή ανάκλησης· η αλυσίδα εκδόσεων διατηρεί το ιστορικό τους.
-Τα vectors επιβιώνουν από ένα crash ή force-kill. Το ευρετήριο vector αποθηκεύεται σε buckets το πολύ κάθε `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 λεπτά). Κάθε vector που προστίθεται ή αφαιρείται στο ενδιάμεσο γράφεται επίσης αμέσως σε ένα μικρό εκκρεμές log στο state store, και η επόμενη εκκίνηση το αναπαράγει χωρίς να καλέσει τον πάροχο embedding. Κάθε επιτυχημένη αποθήκευση αδειάζει το log. Τα έγγραφα που εξακολουθούν να μην έχουν vector μετά την αναπαραγωγή επανα-embeddάρονται στο παρασκήνιο σε batches των `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) μέχρι να μην απομείνει κανένα, και ένα backfill που διακόπτεται συνεχίζει στην επόμενη εκκίνηση. Το `/agentmemory/status` και το viewer δείχνουν το μέγεθος του εκκρεμούς log και την κατάσταση του backfill. Οι εγκαταστάσεις χωρίς κλειδί δεν γράφουν τίποτα.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
Το BM25 κάνει tokenize Ελληνικά, Κυριλλικά, Εβραϊκά, Αραβικά, και τονισμένα Λατινικά απευθείας. Για μνήμες στα Κινεζικά / Ιαπωνικά / Κορεατικά, εγκαταστήστε τους προαιρετικούς segmenters (`npm install @node-rs/jieba tiny-segmenter`) για να χωρίσετε τις ακολουθίες CJK σε tokens επιπέδου λέξης· χωρίς αυτούς, το agentmemory υποχωρεί ομαλά σε tokenization ολόκληρης της ακολουθίας και εκτυπώνει μία υπόδειξη, μία φορά, στο stderr.
@@ -1041,9 +832,9 @@ EMBEDDING_PROVIDER=local
| Πάροχος | Μοντέλο | Κόστος | Σημειώσεις |
|---|---|---|---|
-| **Τοπικό (προτεινόμενη ενεργοποίηση)** | `all-MiniLM-L6-v2` | Δωρεάν | Στη συσκευή μετά την πρώτη λήψη μοντέλου, +8pp ανάκληση σε σχέση με μόνο BM25 |
+| **Τοπικό (προτεινόμενη ενεργοποίηση)** | `all-MiniLM-L6-v2` | Δωρεάν | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Δωρεάν βαθμίδα | 100+ γλώσσες, διαστάσεις 768/1536/3072 (MRL), είσοδος 2048 tokens. Αντικαθιστά το `text-embedding-004` ([καταργήθηκε, διακοπή στις 14 Ιαν 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Υψηλότερη ποιότητα |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Επί πληρωμή | Βελτιστοποιημένο για κώδικα |
| Cohere | `embed-english-v3.0` | Δωρεάν δοκιμή | Γενικής χρήσης |
| OpenRouter | Οποιοδήποτε μοντέλο | Διαφέρει | Proxy πολλών μοντέλων |
diff --git a/READMEs/README.es-ES.md b/READMEs/README.es-ES.md
index a18cec0e..a66ef720 100644
--- a/READMEs/README.es-ES.md
+++ b/READMEs/README.es-ES.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% de recuperación R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% menos tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 tools MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hooks automáticos" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 bases de datos externas" height="38" /></picture>
@@ -331,223 +329,17 @@ Notas de la última release: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Precisión de recuperación
-
-**coding-agent-life-v1** (corpus propio, reproducible en sandbox)
-
-| Adaptador | P@5 | R@5 | Tasa de aciertos top-5 | Latencia p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| baseline grep | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Tasa de aciertos top-5 del 100% en el **techo matemático de P@5** para este corpus (0.240, ver scorecard). Hybrid recupera todas las sesiones gold; grep falla 1 de 2 gold en la consulta temporal multi-sesión. La mejora es **recall + temporal**, no precisión agregada. Este benchmark es pequeño y escaso en gold; el LongMemEval-S más grande de abajo diferencia mejor. Desglose completo por tipo + nota de corrección: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 preguntas)
+Resultados actuales de recuperación, definiciones de métricas y pasos para reproducirlos:
-| Sistema | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Fallback solo BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Ahorro de tokens
-
-| Enfoque | Tokens/año | Coste/año |
-|---|---|---|
-| Pegar el contexto completo | 19.5M+ | Imposible (excede la ventana) |
-| Resumido por LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embeddings locales | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Modelo de embedding: `all-MiniLM-L6-v2` (local, gratis, sin API key). Informes completos: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Comparativa con competidores: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), que cubre agentmemory frente a mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reprodúcelo en local:** [`eval/README.md`](../eval/README.md), un harness con adaptadores intercambiables para LongMemEval `_s` (500 preguntas públicas) + `coding-agent-life-v1` (corpus propio de 15 sesiones). Los adaptadores grep / vector / agentmemory puntúan uno junto al otro, salida NDJSON, los scorecards publicados aterrizan en [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Funciona muy bien con [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) y [Graphify](https://github.com/safishamsi/graphify).** Indexado de grafos de código, pipelines de build multiagente y grafos de conocimiento más amplios sobre documentos / PDFs / imágenes / vídeos. agentmemory recuerda el trabajo; esos tres proyectos iluminan el resto de la capa de contexto. Recetas y tabla de enrutamiento por pregunta: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Comparativa" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Integrada (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tipo</strong></td>
-<td>Motor de memoria + servidor MCP</td>
-<td>API de capa de memoria</td>
-<td>Runtime de agente completo</td>
-<td>IA personal</td>
-<td>API de memoria + app</td>
-<td>Hub de memoria de equipo (proxy LLM)</td>
-<td>Memoria vectorial (OSS)</td>
-<td>Motor de memoria (Oracle DB)</td>
-<td>Sistema de memoria</td>
-<td>Fichero estático</td>
-</tr>
-<tr>
-<td><strong>R@5 de recuperación</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Autoinformado</td>
-<td>PersonaMem 76% (autoinformado)</td>
-<td>~96.6% (autoinformado)</td>
-<td>94.4% (autoinformado)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Captura automática</strong></td>
-<td>12 hooks (esfuerzo manual cero)</td>
-<td>Llamadas manuales a <code>add()</code></td>
-<td>Autoedición por el agente</td>
-<td>Manual</td>
-<td>Extracción del lado de la API</td>
-<td>Intercepción por proxy (cambio de base-URL)</td>
-<td>Manual</td>
-<td>Extracción por API</td>
-<td>Manual</td>
-<td>Edición manual</td>
-</tr>
-<tr>
-<td><strong>Búsqueda</strong></td>
-<td>BM25 + Vector + Graph (fusión RRF)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semántica</td>
-<td>Vector + RAG</td>
-<td>4 tipos de asset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Solo vector</td>
-<td>Vector + semántica</td>
-<td>Ponderada por decaimiento</td>
-<td>Carga todo en el contexto</td>
-</tr>
-<tr>
-<td><strong>Multiagente</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (sin coordinación)</td>
-<td>Solo dentro del runtime de Letta</td>
-<td>No</td>
-<td>No</td>
-<td>Roles de equipo + assets compartidos</td>
-<td>No</td>
-<td>Solo con scope</td>
-<td>Compartido multiagente</td>
-<td>Ficheros por agente</td>
-</tr>
-<tr>
-<td><strong>Lock-in de framework</strong></td>
-<td>Ninguno (cualquier cliente MCP)</td>
-<td>Ninguno</td>
-<td>Alto (debes usar Letta)</td>
-<td>Standalone</td>
-<td>Ninguno</td>
-<td>Un proxy antepone cada llamada al modelo</td>
-<td>Ninguno</td>
-<td>Oracle Database</td>
-<td>Ninguno</td>
-<td>Formato por agente</td>
-</tr>
-<tr>
-<td><strong>Dependencias externas</strong></td>
-<td>No necesita un servicio de base de datos separado</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + BD vectorial</td>
-<td>Varias</td>
-<td>Nube gestionada</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Ninguna</td>
-<td>Ninguna</td>
-</tr>
-<tr>
-<td><strong>Ciclo de vida de la memoria</strong></td>
-<td>Consolidación de 4 niveles + decaimiento + auto-olvido</td>
-<td>Extracción pasiva</td>
-<td>Gestionado por el agente</td>
-<td>Manual</td>
-<td>Auto-olvido</td>
-<td>Revisión manual; auto-enrutado en progreso</td>
-<td>Ninguno</td>
-<td>No especificado</td>
-<td>Decaimiento + consolidación</td>
-<td>Poda manual</td>
-</tr>
-<tr>
-<td><strong>Eficiencia de tokens</strong></td>
-<td>~1,900 tokens/sesión ($10/año)</td>
-<td>Varía según la integración</td>
-<td>Memoria core en el contexto</td>
-<td>Varía</td>
-<td>Precios de nube</td>
-<td>No especificado</td>
-<td>Sin presupuesto de tokens</td>
-<td>Respaldado por LLM (varía)</td>
-<td>Varía</td>
-<td>22K+ tokens con 240 obs</td>
-</tr>
-<tr>
-<td><strong>Visor en tiempo real</strong></td>
-<td>Sí (puerto 3113)</td>
-<td>Dashboard en la nube</td>
-<td>Dashboard en la nube</td>
-<td>UI web</td>
-<td>Dashboard en la nube</td>
-<td>UI web del hub</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Sí (por defecto)</td>
-<td>Opcional</td>
-<td>Opcional</td>
-<td>Sí</td>
-<td>No (solo nube)</td>
-<td>Sí (Docker)</td>
-<td>Sí</td>
-<td>Sí (Oracle DB)</td>
-<td>Sí</td>
-<td>Sí</td>
-</tr>
-</table>
-
-<sub>Nota de benchmark: solo el R@5 de agentmemory es un resultado medido por nosotros (LongMemEval-S, reproducible desde <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Las cifras de mem0 y Letta son sus números publicados de LoCoMo (un dataset distinto); las cifras de MemPalace, supermemory, TencentDB (PersonaMem) y oracleagentmemory son afirmaciones autoinformadas por el proveedor que no hemos reproducido de forma independiente (la ejecución de oracleagentmemory usó GPT-5.5 contra una Oracle AI Database). Se muestran una junto a otra solo a modo orientativo, no como un cara a cara sobre datos idénticos. Los conteos de estrellas son aproximados y varían con el tiempo.</sub>
-
-**Nuevos entrantes** que vale la pena conocer, comparados en profundidad en [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistema | ⭐ | Enfoque |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Grafo de conocimiento temporal; los resultados publicados más fuertes en consultas temporales (LongMemEval 63.8%), pero el grafo se construye de forma asíncrona, así que los hechos recientes pueden tardar en aparecer |
-| Cognee | 30K | Ingesta de documento a grafo de conocimiento, solo Python, construido para extracción estructurada de entidades en lugar de captura de sesiones |
+Comparación actualizada de productos y opciones de despliegue, con fuentes:
-AgentMemory combina captura desde agentes de programación, historial de proyecto consultable y un visor local. Puedes usarlo sin clave desde el primer inicio.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Guía completa: [`integrations/hermes/`](../integrations/hermes/)
Arranca el servidor de memoria: `npx -y @agentmemory/agentmemory@latest`
-#### Skills nativas vía `npx skills add` (50+ agentes)
+#### Skills nativas vía `npx skills add`
agentmemory incluye 17 skills en el formato `<dir>/SKILL.md` al estilo de Claude Code: 9 skills de acción invocables (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) y 8 skills de referencia que el agente carga bajo demanda (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Las skills de referencia llevan tablas de datos generadas desde el código fuente, así que nunca se desincronizan. La CLI [`skills`](https://npmjs.com/package/skills) de vercel-labs las auto-instala en el directorio de skills nativo del agente que la invoca en 50+ agentes (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf y más):
@@ -836,7 +628,7 @@ La entrada de agentmemory es el **mismo bloque de servidor MCP** en todos los ho
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` instala el MCP y los hooks de captura en el directorio de personalización compartido. Consulta [configuración y límites de Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` usa la misma configuración de MCP y hooks que las versiones actuales del IDE. Las instalaciones existentes deben actualizar con `--force`; consulta las [notas de actualización](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` escribe la configuración a nivel de usuario. Los overrides de workspace van en `.kiro/settings/mcp.json` junto a tu código. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escribe el bloque `mcpServers` estándar. Warp también auto-descubre skills desde `.claude/skills/`; una vez instalado el plugin de Claude Code, las 8 skills de agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) aparecen de forma nativa en la paleta de slash commands de Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escribe el bloque `mcpServers` estándar. Warp también auto-descubre skills desde `.claude/skills/`; una vez instalado el plugin de Claude Code, las 9 skills de agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) aparecen de forma nativa en la paleta de slash commands de Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` escribe el bloque `mcpServers` estándar. Usuarios de la extensión de VS Code: pega el mismo bloque vía Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferido) o `config.json` (legacy) | `agentmemory connect continue` crea `config.yaml` si falta o modifica el `config.json` existente. Si ya existe YAML, muestra el bloque que debes añadir a `mcpServers:` para conservar comentarios y anclas. Continue usa un array, no un objeto. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` escribe bajo `context_servers` (la clave de Zed, NO `mcpServers`). Los servidores MCP remotos se pueden conectar en su lugar vía `{"url": "..."}`. |
@@ -927,7 +719,6 @@ Todo agente de codificación con IA viene con memoria integrada: Claude Code tie
|---|---|---|
| Mantenimiento | Instrucciones mantenidas manualmente | Historial de sesiones y recuerdos indexados |
| Búsqueda | Carga todo en contexto | BM25 + vector + graph (solo top-K) |
-| Coste en tokens | 22K+ con 240 observaciones | ~1,900 tokens (92% menos) |
| Cross-agent | Ficheros por agente | MCP + REST (cualquier agente) |
| Coordinación | Ninguna | Leases, signals, actions, routines |
| Observabilidad | Lectura manual de ficheros | Visor en tiempo real en :3113 |
@@ -940,7 +731,7 @@ Todo agente de codificación con IA viene con memoria integrada: Claude Code tie
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ Fusionado con Reciprocal Rank Fusion (RRF, k=60) y diversificado por sesión (m
Cuando el índice vectorial está poblado, `mem::search` (detrás de `memory_recall`) usa el ranker híbrido BM25 + vector. Sin embeddings usa BM25. `smart-search` además puede fusionar coincidencias estructurales del grafo cuando existen datos de grafo, incluso en modo sin claves. El recall de lecciones corre sobre un índice BM25 in-memory dedicado en lugar de escanear todo el corpus en cada consulta. Las versiones de memoria reemplazadas quedan excluidas de todas las rutas de recall; la cadena de versiones conserva su historial.
-Los vectores sobreviven a un crash o un force-kill. El índice vectorial se guarda en buckets como máximo cada `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutos). Cada vector añadido o eliminado entretanto también se escribe de inmediato en un pequeño log pendiente en el state store, y el siguiente arranque lo reproduce sin llamar al proveedor de embeddings. Cada guardado exitoso vacía el log. Los documentos que siguen sin vector tras la reproducción se re-embeben en segundo plano en lotes de `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) hasta que no quede ninguno, y un backfill que se detiene continúa en el siguiente arranque. `/agentmemory/status` y el visor muestran el tamaño del log pendiente y el estado del backfill. Las instalaciones sin claves no escriben nada.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokeniza griego, cirílico, hebreo, árabe y latín con tildes de serie. Para memorias en chino / japonés / coreano, instala los segmentadores opcionales (`npm install @node-rs/jieba tiny-segmenter`) para partir los runs CJK en tokens a nivel de palabra; sin ellos, agentmemory hace soft-fallback a tokenización por run completo y muestra una pista única en stderr.
@@ -1041,9 +832,9 @@ La instalación npm normal incluye el runtime opcional `@huggingface/transformer
| Proveedor | Modelo | Coste | Notas |
|---|---|---|---|
-| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratis | En local tras la primera descarga del modelo, +8pp de recall sobre BM25-only |
+| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratis | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free tier | 100+ idiomas, 768/1536/3072 dims (MRL), entrada de 2048 tokens. Sustituye a `text-embedding-004` ([deprecado, cierre el 14 ene 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Máxima calidad |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | De pago | Optimizado para código |
| Cohere | `embed-english-v3.0` | Trial gratis | Uso general |
| OpenRouter | Cualquier modelo | Varía | Proxy multi-modelo |
diff --git a/READMEs/README.fi-FI.md b/READMEs/README.fi-FI.md
index 02dab56f..24c7d170 100644
--- a/READMEs/README.fi-FI.md
+++ b/READMEs/README.fi-FI.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95,2 % hakutarkkuus R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92 % vähemmän tokeneita" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP-työkalua" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automaattista hookia" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 ulkoista tietokantaa" height="38" /></picture>
@@ -333,223 +331,17 @@ Uusimmat julkaisutiedot: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarkit" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Hakutarkkuus
-
-**coding-agent-life-v1** (talon sisäinen korpus, hiekkalaatikossa toistettavissa)
-
-| Sovitin | P@5 | R@5 | Top-5-osumasuhde | p50-latenssi |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-perustaso | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100 % top-5-osumasuhde **P@5-matemaattisessa kattossa** tälle korpukselle (0.240, katso scorecard). Hybrid hakee jokaisen kultaisen istunnon. grep jättää huomiotta 1/2 kultaisesta moni-istuntoisessa temporaalikyselyssä. Hyöty on **muistaminen + temporaalisuus**, ei kokonaistarkkuus. Tämä benchmark on pieni ja kultadataltaan harva; isompi LongMemEval-S alla erottelee paremmin. Täysi tyyppikohtainen erittely + korjaushuomautus: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 kysymystä)
+Ajantasaiset hakutulokset, mittarien määritelmät ja toisto-ohjeet:
-| Järjestelmä | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| vain BM25 -varmistusreitti | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tokenisäästöt
-
-| Lähestymistapa | Tokenia/vuosi | Kustannus/vuosi |
-|---|---|---|
-| Liitä koko konteksti | 19.5M+ | Mahdotonta (ylittää ikkunan) |
-| LLM-tiivistetty | ~650K | ~500 $ |
-| **agentmemory** | **~170K** | **~10 $** |
-| agentmemory + paikalliset upotukset | ~170K | **0 $** |
-
-</td>
-</tr>
-</table>
-
-> Upotusmalli: `all-MiniLM-L6-v2` (paikallinen, ilmainen, ei API-avainta). Täydet raportit: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Kilpailijavertailu: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), joka kattaa agentmemoryn vs. mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Toista paikallisesti:** [`eval/README.md`](../eval/README.md), sovitinpohjainen testipenkki LongMemEvalille `_s` (julkinen 500 kysymystä) + `coding-agent-life-v1` (talon sisäinen 15 istunnon korpus). Grep-, vektori- ja agentmemory-sovittimet pisteytetään rinnakkain, NDJSON-tulostus, julkaistut scorecardit löytyvät kansiosta [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Toimii yhdessä [codegraphin](https://github.com/colbymchenry/codegraph), [Understand Anythingin](https://github.com/Lum1104/Understand-Anything) ja [Graphifyn](https://github.com/safishamsi/graphify) kanssa.** Koodigraafi-indeksointi, moniagenttiset build-putket ja laajemmat tietograafit dokumenttien / PDF:ien / kuvien / videoiden yli. agentmemory muistaa työn; nämä kolme projektia valaisevat kontekstikerroksen muun osan. Reseptit + kysymysten reititystaulukko: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs. kilpailijat" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Sisäänrakennettu (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tyyppi</strong></td>
-<td>Muistimoottori + MCP-palvelin</td>
-<td>Muistikerros-API</td>
-<td>Täysi agentti-ajonaikainen ympäristö</td>
-<td>Henkilökohtainen AI</td>
-<td>Muisti-API + sovellus</td>
-<td>Tiimimuistikeskus (LLM-proxy)</td>
-<td>Vektorimuisti (avoin lähdekoodi)</td>
-<td>Muistimoottori (Oracle-tietokanta)</td>
-<td>Muistijärjestelmä</td>
-<td>Staattinen tiedosto</td>
-</tr>
-<tr>
-<td><strong>Hakutarkkuus R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>Ei saatavilla</td>
-<td>Itse ilmoitettu</td>
-<td>PersonaMem 76% (itse ilmoitettu)</td>
-<td>~96.6% (itse ilmoitettu)</td>
-<td>94.4% (itse ilmoitettu)</td>
-<td>Ei saatavilla</td>
-<td>Ei saatavilla (grep)</td>
-</tr>
-<tr>
-<td><strong>Automaattinen tallennus</strong></td>
-<td>12 hookia (ei manuaalista työtä)</td>
-<td>Manuaaliset <code>add()</code>-kutsut</td>
-<td>Agentti muokkaa itseään</td>
-<td>Manuaalinen</td>
-<td>API-puolen poiminta</td>
-<td>Proxy-sieppaus (base-URL-vaihto)</td>
-<td>Manuaalinen</td>
-<td>API-poiminta</td>
-<td>Manuaalinen</td>
-<td>Manuaalinen muokkaus</td>
-</tr>
-<tr>
-<td><strong>Haku</strong></td>
-<td>BM25 + vektori + graafi (RRF-fuusio)</td>
-<td>Vektori + graafi</td>
-<td>Vektori (arkisto)</td>
-<td>Semanttinen</td>
-<td>Vektori + RAG</td>
-<td>4 omaisuustyyppiä (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vain vektori</td>
-<td>Vektori + semanttinen</td>
-<td>Vanhenemispainotettu</td>
-<td>Lataa kaiken kontekstiin</td>
-</tr>
-<tr>
-<td><strong>Moniagenttisuus</strong></td>
-<td>MCP + REST + leaset + signaalit</td>
-<td>API (ei koordinaatiota)</td>
-<td>Vain Lettan ajonaikaisessa ympäristössä</td>
-<td>Ei</td>
-<td>Ei</td>
-<td>Tiimiroolit + jaetut omaisuudet</td>
-<td>Ei</td>
-<td>Vain rajattu</td>
-<td>Jaettu moniagenttisuus</td>
-<td>Agenttikohtaiset tiedostot</td>
-</tr>
-<tr>
-<td><strong>Kehyksestä riippuvuus</strong></td>
-<td>Ei lainkaan (mikä tahansa MCP-asiakas)</td>
-<td>Ei lainkaan</td>
-<td>Korkea (pakko käyttää Lettaa)</td>
-<td>Itsenäinen</td>
-<td>Ei lainkaan</td>
-<td>Proxy edessä joka mallikutsussa</td>
-<td>Ei lainkaan</td>
-<td>Oracle Database</td>
-<td>Ei lainkaan</td>
-<td>Agenttikohtainen formaatti</td>
-</tr>
-<tr>
-<td><strong>Ulkoiset riippuvuudet</strong></td>
-<td>Erillistä tietokantapalvelua ei tarvita</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektoritietokanta</td>
-<td>Useita</td>
-<td>Hallittu pilvi</td>
-<td>Docker-pino (Core + Hub + Proxy)</td>
-<td>Vektorivarasto</td>
-<td>Oracle AI Database</td>
-<td>Ei mitään</td>
-<td>Ei mitään</td>
-</tr>
-<tr>
-<td><strong>Muistin elinkaari</strong></td>
-<td>4-tasoinen konsolidointi + vanheneminen + automaattinen unohtaminen</td>
-<td>Passiivinen poiminta</td>
-<td>Agentin hallinnoima</td>
-<td>Manuaalinen</td>
-<td>Automaattinen unohtaminen</td>
-<td>Manuaalinen katselmointi; automaattireititys kehitteillä</td>
-<td>Ei mitään</td>
-<td>Ei ilmoitettu</td>
-<td>Vanheneminen + konsolidointi</td>
-<td>Manuaalinen karsinta</td>
-</tr>
-<tr>
-<td><strong>Tokentehokkuus</strong></td>
-<td>~1 900 tokenia/istunto (10 $/vuosi)</td>
-<td>Vaihtelee integraation mukaan</td>
-<td>Ydinmuisti kontekstissa</td>
-<td>Vaihtelee</td>
-<td>Pilvihinnoittelu</td>
-<td>Ei ilmoitettu</td>
-<td>Ei tokenbudjettia</td>
-<td>LLM-pohjainen (vaihtelee)</td>
-<td>Vaihtelee</td>
-<td>22K+ tokenia 240 havainnolla</td>
-</tr>
-<tr>
-<td><strong>Reaaliaikainen katseluohjelma</strong></td>
-<td>Kyllä (portti 3113)</td>
-<td>Pilvikojelauta</td>
-<td>Pilvikojelauta</td>
-<td>Web-käyttöliittymä</td>
-<td>Pilvikojelauta</td>
-<td>Hub-web-käyttöliittymä</td>
-<td>Ei</td>
-<td>Ei</td>
-<td>Ei</td>
-<td>Ei</td>
-</tr>
-<tr>
-<td><strong>Itseisännöity</strong></td>
-<td>Kyllä (oletus)</td>
-<td>Valinnainen</td>
-<td>Valinnainen</td>
-<td>Kyllä</td>
-<td>Ei (vain pilvi)</td>
-<td>Kyllä (Docker)</td>
-<td>Kyllä</td>
-<td>Kyllä (Oracle-tietokanta)</td>
-<td>Kyllä</td>
-<td>Kyllä</td>
-</tr>
-</table>
-
-<sub>Benchmark-huomautus: vain agentmemoryn R@5 on omaa mitattua tulostamme (LongMemEval-S, toistettavissa tiedostosta <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). mem0- ja Letta-luvut ovat niiden julkaisemia LoCoMo-lukuja (eri aineisto); MemPalace-, supermemory-, TencentDB (PersonaMem) - ja oracleagentmemory-luvut ovat valmistajien itse ilmoittamia väitteitä, joita emme ole itsenäisesti toistaneet (oracleagentmemoryn ajo käytti GPT-5.5:tä Oracle AI Databasea vastaan). Esitetty rinnakkain vain suuntaa antavana, ei pää-pää-vertailuna samalla datalla. Tähtimäärät ovat likimääräisiä ja muuttuvat ajan myötä.</sub>
-
-**Uudemmat tulokkaat**, hyvä tietää, syvemmin vertailtuna kohdassa [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Järjestelmä | ⭐ | Näkökulma |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporaalinen tietograafi; vahvimmat julkaistut temporaalikyselytulokset (LongMemEval 63.8%), mutta graafi rakentuu asynkronisesti, jolloin tuoreet faktat voivat viivästyä |
-| Cognee | 30K | Dokumentista tietograafiksi -tuonti, vain Python, rakennettu rakenteelliseen entiteettipoimintaan istuntotallennuksen sijaan |
+Päivitetty tuotteiden ja käyttöönottovaihtoehtojen vertailu lähteineen:
-AgentMemory yhdistää istuntojen tallennuksen, haettavan projektihistorian ja paikallisen viewerin; avaimeton käyttö toimii ensimmäisestä käynnistyksestä.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Täysi opas: [`integrations/hermes/`](../integrations/hermes/)
Käynnistä muistipalvelin: `npx -y @agentmemory/agentmemory@latest`
-#### Natiivit taidot komennolla `npx skills add` (50+ agenttia)
+#### Natiivit taidot komennolla `npx skills add`
agentmemory toimittaa 17 taitoa Claude-Code-tyylisessä `<dir>/SKILL.md`-formaatissa: 9 kutsuttavaa toimintotaitoa (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) ja 8 viitetaitoa, jotka agentti lataa tarpeen mukaan (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Viitetaidot kuljettavat lähdekoodista generoituja datataulukoita, niin ne eivät koskaan ajaudu pois synkasta. vercel-labsin [`skills`](https://npmjs.com/package/skills)-CLI asentaa ne automaattisesti kutsuvan agentin natiiviin taitohakemistoon yli 50 agentin yli (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf ja monet muut):
@@ -844,7 +636,7 @@ agentmemory-merkintä on **sama MCP-palvelinlohko** jokaisessa isännässä, jok
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` asentaa MCP:n ja tallennushookit jaettuun mukautushakemistoon. Katso [Antigravityn asennus ja rajoitukset](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` käyttää samaa MCP- ja hook-määritystä kuin nykyiset IDE-versiot. Olemassa olevat asennukset tulisi päivittää `--force`-valitsimella; katso [päivitysohjeet](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` kirjoittaa käyttäjätason asetukset. Työtilakohtaiset ylikirjoitukset menevät tiedostoon `.kiro/settings/mcp.json` koodisi vierelle. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` kirjoittaa standardin `mcpServers`-lohkon. Warp myös tunnistaa automaattisesti taidot kansiosta `.claude/skills/`; kun Claude Code -plugin on asennettu, 8 agentmemory-taitoa (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) näkyvät natiivisti Warpin pikakomentopaletissa. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` kirjoittaa standardin `mcpServers`-lohkon. Warp myös tunnistaa automaattisesti taidot kansiosta `.claude/skills/`; kun Claude Code -plugin on asennettu, 9 agentmemory-taitoa (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) näkyvät natiivisti Warpin pikakomentopaletissa. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` kirjoittaa standardin `mcpServers`-lohkon. VS Code -laajennuksen käyttäjät: liitä sama lohko Cline Settings → MCP Servers → Edit JSON -kautta. |
| **Continue.dev** | `~/.continue/config.yaml` (suositeltu) tai `config.json` (vanha) | `agentmemory connect continue` luo `config.yaml`-tiedoston, jos asetuksia ei ole, tai muokkaa olemassa olevaa `config.json`-tiedostoa. Olemassa olevan YAML:n tapauksessa se tulostaa lisättävän `mcpServers:`-lohkon säilyttääkseen kommentit ja ankkurit. Continue käyttää taulukkoa, ei objektia. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` kirjoittaa avaimen `context_servers` alle (Zedin avain, EI `mcpServers`). Etä-MCP-palvelimet voidaan kytkeä myös muodolla `{"url": "..."}`. |
@@ -927,7 +719,6 @@ Jokainen tekoälypohjainen koodausagentti toimitetaan sisäänrakennetulla muist
|---|---|---|
| Ylläpito | Käsin ylläpidetyt ohjeet | Indeksoitu istuntohistoria ja muistot |
| Haku | Lataa kaiken kontekstiin | BM25 + vektori + graafi (vain top-K) |
-| Tokenkustannus | 22K+ 240 havainnolla | ~1 900 tokenia (92 % vähemmän) |
| Agenttien välillä | Agenttikohtaiset tiedostot | MCP + REST (mikä tahansa agentti) |
| Koordinointi | Ei mitään | Leaset, signaalit, toiminnot, rutiinit |
| Havainnoitavuus | Lue tiedostoja käsin | Reaaliaikainen katseluohjelma portissa :3113 |
@@ -940,7 +731,7 @@ Jokainen tekoälypohjainen koodausagentti toimitetaan sisäänrakennetulla muist
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ Yhdistetty Reciprocal Rank Fusionilla (RRF, k=60) ja istuntokohtaisesti monipuol
Kun vektori-indeksi on täytetty, `mem::search` (`memory_recall`-työkalun takana) käyttää hybridiä BM25 + vektori -luokittelijaa. Ilman upotuksia se käyttää BM25:tä. `smart-search` voi lisäksi yhdistää rakenteellisia graafiosumia, kun graafidataa on olemassa, myös avaimettomassa tilassa. Opetusten palautus käyttää erillistä muistinvaraista BM25-indeksiä sen sijaan, että se skannaisi koko korpuksen jokaisella kyselyllä. Korvatut muistiversiot on suljettu pois jokaisesta palautusreitistä; versioketju säilyttää niiden historian.
-Vektorit selviävät kaatumisesta tai pakkosammutuksesta. Vektori-indeksi tallennetaan erissä enintään `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`-ajan välein (10 minuuttia). Joka vektori, joka lisätään tai poistetaan tällä välillä, kirjoitetaan myös heti pieneen odottavaan lokiin tilavarastossa, ja seuraava käynnistys toistaa sen kutsumatta upotustarjoajaa. Joka onnistunut tallennus tyhjentää lokin. Dokumentit, joilla ei vielä ole vektoria toiston jälkeen, upotetaan uudelleen taustalla erissä kooltaan `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), kunnes yhtäkään ei jää jäljelle, ja pysäytetty täydennys jatkuu seuraavassa käynnistyksessä. `/agentmemory/status` ja katseluohjelma näyttävät odottavan lokin koon ja täydennyksen tilan. Avaimettomat asennukset eivät kirjoita mitään.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenisoi kreikan, kyrilliset kirjaimet, hebrean, arabian ja aksentoidun latinan oletuksena. Kiinan-, japanin- ja koreankielisiä muistoja varten asenna valinnaiset segmentoijat (`npm install @node-rs/jieba tiny-segmenter`) jakaaksesi CJK-jaksot sananlaajuisiksi tokeneiksi; ilman niitä agentmemory lankeaa pehmeästi koko jakson tokenisointiin ja tulostaa kertaluonteisen vihjeen stderriin.
@@ -1041,9 +832,9 @@ Normaali npm-asennus sisältää valinnaisen `@huggingface/transformers`-ajonaik
| Tarjoaja | Malli | Kustannus | Huomautuksia |
|---|---|---|---|
-| **Paikallinen (suositeltu opt-in)** | `all-MiniLM-L6-v2` | Ilmainen | Laitteella ensimmäisen mallilatauksen jälkeen, +8pp palautustarkkuutta vain-BM25:een verrattuna |
+| **Paikallinen (suositeltu opt-in)** | `all-MiniLM-L6-v2` | Ilmainen | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Ilmainen taso | 100+ kieltä, 768/1536/3072-ulotteinen (MRL), 2048-tokenin syöte. Korvaa `text-embedding-004`:n ([poistettu käytöstä, sammutetaan 14.1.2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | 0,02 $/1M | Korkein laatu |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Maksullinen | Optimoitu koodille |
| Cohere | `embed-english-v3.0` | Ilmainen kokeilu | Yleiskäyttöinen |
| OpenRouter | Mikä tahansa malli | Vaihtelee | Multi-model-proxy |
diff --git a/READMEs/README.fr-FR.md b/READMEs/README.fr-FR.md
index c0ef2f9d..e7172efa 100644
--- a/READMEs/README.fr-FR.md
+++ b/READMEs/README.fr-FR.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -331,223 +329,17 @@ Notes de la dernière version : [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Précision de récupération
-
-**coding-agent-life-v1** (corpus interne, reproductible en sandbox)
-
-| Adaptateur | P@5 | R@5 | Taux de hit top-5 | Latence p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| Référence grep | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% de taux de hit top-5 au **plafond mathématique du P@5** pour ce corpus (0.240, voir le scorecard). L'hybride récupère chaque session gold ; grep manque 1 des 2 gold sur la requête temporelle multi-session. Le gain porte sur **recall + temporel**, pas sur la précision agrégée. Ce benchmark est petit et pauvre en gold ; le LongMemEval-S plus grand ci-dessous différencie mieux. Ventilation complète par type + note de correction : [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 questions)
+Résultats actuels de recherche, définitions des métriques et étapes de reproduction :
-| Système | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Repli BM25 seul | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Économies de tokens
-
-| Approche | Tokens/an | Coût/an |
-|---|---|---|
-| Coller le contexte complet | 19,5M+ | Impossible (dépasse la fenêtre) |
-| Résumé par LLM | ~650K | ~500 $ |
-| **agentmemory** | **~170K** | **~10 $** |
-| agentmemory + embeddings locaux | ~170K | **0 $** |
-
-</td>
-</tr>
-</table>
-
-> Modèle d'embedding : `all-MiniLM-L6-v2` (local, gratuit, sans clé d'API). Rapports complets : [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Comparaison avec les concurrents : [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) couvrant agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproduire localement :** [`eval/README.md`](../eval/README.md), un harnais à adaptateurs pluggables pour LongMemEval `_s` (public, 500 questions) + `coding-agent-life-v1` (corpus interne de 15 sessions). Les adaptateurs grep / vectoriel / agentmemory sont notés côte à côte, sortie NDJSON, scorecards publiés dans [`docs/benchmarks/`](../docs/benchmarks/).
-
-**À associer à [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) et [Graphify](https://github.com/safishamsi/graphify).** Indexation de graphe de code, pipelines de build multi-agents et graphes de connaissances plus larges sur docs / PDF / images / vidéos. agentmemory mémorise le travail ; ces trois projets éclairent le reste de la couche de contexte. Recettes + tableau de routage des questions : [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Concurrents" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Intégré (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Type</strong></td>
-<td>Moteur de mémoire + serveur MCP</td>
-<td>API de couche mémoire</td>
-<td>Runtime d'agent complet</td>
-<td>IA personnelle</td>
-<td>API mémoire + app</td>
-<td>Hub de mémoire d'équipe (proxy LLM)</td>
-<td>Mémoire vectorielle (OSS)</td>
-<td>Moteur de mémoire (Oracle DB)</td>
-<td>Système de mémoire</td>
-<td>Fichier statique</td>
-</tr>
-<tr>
-<td><strong>R@5 de récupération</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Auto-déclaré</td>
-<td>PersonaMem 76% (auto-déclaré)</td>
-<td>~96.6% (auto-déclaré)</td>
-<td>94.4% (auto-déclaré)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Capture automatique</strong></td>
-<td>12 hooks (zéro effort manuel)</td>
-<td>Appels <code>add()</code> manuels</td>
-<td>L'agent s'auto-édite</td>
-<td>Manuelle</td>
-<td>Extraction côté API</td>
-<td>Interception par proxy (bascule de base-URL)</td>
-<td>Manuelle</td>
-<td>Extraction API</td>
-<td>Manuelle</td>
-<td>Édition manuelle</td>
-</tr>
-<tr>
-<td><strong>Recherche</strong></td>
-<td>BM25 + Vectoriel + Graphe (fusion RRF)</td>
-<td>Vectoriel + Graphe</td>
-<td>Vectoriel (archival)</td>
-<td>Sémantique</td>
-<td>Vectoriel + RAG</td>
-<td>4 types d'assets (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vectoriel uniquement</td>
-<td>Vectoriel + sémantique</td>
-<td>Pondérée par décroissance</td>
-<td>Charge tout en contexte</td>
-</tr>
-<tr>
-<td><strong>Multi-agents</strong></td>
-<td>MCP + REST + leases + signaux</td>
-<td>API (sans coordination)</td>
-<td>Uniquement dans le runtime Letta</td>
-<td>Non</td>
-<td>Non</td>
-<td>Rôles d'équipe + assets partagés</td>
-<td>Non</td>
-<td>Scopé seulement</td>
-<td>Partagé multi-agents</td>
-<td>Fichiers par agent</td>
-</tr>
-<tr>
-<td><strong>Verrouillage framework</strong></td>
-<td>Aucun (tout client MCP)</td>
-<td>Aucun</td>
-<td>Élevé (Letta obligatoire)</td>
-<td>Autonome</td>
-<td>Aucun</td>
-<td>Le proxy s'interpose devant chaque appel de modèle</td>
-<td>Aucun</td>
-<td>Oracle Database</td>
-<td>Aucun</td>
-<td>Format par agent</td>
-</tr>
-<tr>
-<td><strong>Dépendances externes</strong></td>
-<td>Aucun service de base de données séparé requis</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + base vectorielle</td>
-<td>Multiples</td>
-<td>Cloud managé</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Store vectoriel</td>
-<td>Oracle AI Database</td>
-<td>Aucune</td>
-<td>Aucune</td>
-</tr>
-<tr>
-<td><strong>Cycle de vie mémoire</strong></td>
-<td>Consolidation à 4 niveaux + décroissance + oubli automatique</td>
-<td>Extraction passive</td>
-<td>Géré par l'agent</td>
-<td>Manuel</td>
-<td>Oubli automatique</td>
-<td>Revue manuelle ; auto-routage en cours</td>
-<td>Aucun</td>
-<td>Non précisé</td>
-<td>Décroissance + consolidation</td>
-<td>Nettoyage manuel</td>
-</tr>
-<tr>
-<td><strong>Efficacité des tokens</strong></td>
-<td>~1 900 tokens/session (10 $/an)</td>
-<td>Varie selon l'intégration</td>
-<td>Mémoire centrale dans le contexte</td>
-<td>Varie</td>
-<td>Tarification cloud</td>
-<td>Non précisé</td>
-<td>Pas de budget de tokens</td>
-<td>Adossé à un LLM (variable)</td>
-<td>Varie</td>
-<td>22K+ tokens à 240 observations</td>
-</tr>
-<tr>
-<td><strong>Visualiseur temps réel</strong></td>
-<td>Oui (port 3113)</td>
-<td>Dashboard cloud</td>
-<td>Dashboard cloud</td>
-<td>UI web</td>
-<td>Dashboard cloud</td>
-<td>UI web du Hub</td>
-<td>Non</td>
-<td>Non</td>
-<td>Non</td>
-<td>Non</td>
-</tr>
-<tr>
-<td><strong>Auto-hébergé</strong></td>
-<td>Oui (par défaut)</td>
-<td>Optionnel</td>
-<td>Optionnel</td>
-<td>Oui</td>
-<td>Non (cloud uniquement)</td>
-<td>Oui (Docker)</td>
-<td>Oui</td>
-<td>Oui (Oracle DB)</td>
-<td>Oui</td>
-<td>Oui</td>
-</tr>
-</table>
-
-<sub>Note sur le benchmark : seul le R@5 d'agentmemory est notre propre résultat mesuré (LongMemEval-S, reproductible depuis <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Les chiffres de mem0 et Letta sont leurs résultats LoCoMo publiés (un jeu de données différent) ; les chiffres de MemPalace, supermemory, TencentDB (PersonaMem) et oracleagentmemory sont des annonces auto-déclarées par ces éditeurs, que nous n'avons pas reproduites indépendamment (le run d'oracleagentmemory utilisait GPT-5.5 contre une Oracle AI Database). Présentés côte à côte à titre indicatif seulement, pas comme un comparatif direct sur des données identiques. Les comptages d'étoiles sont approximatifs et évoluent avec le temps.</sub>
-
-**Nouveaux arrivants** à connaître, comparés en détail dans [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) :
-
-| Système | ⭐ | Angle |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Graphe de connaissances temporel ; meilleurs résultats publiés sur les requêtes temporelles (LongMemEval 63.8%), mais le graphe se construit de façon asynchrone donc les faits récents peuvent accuser un retard |
-| Cognee | 30K | Ingestion document-vers-graphe-de-connaissances, Python uniquement, conçu pour l'extraction d'entités structurées plutôt que pour la capture de sessions |
+Comparaison actualisée des produits et des options de déploiement, avec sources :
-AgentMemory associe capture par les agents de programmation, historique de projet interrogeable et visualiseur local. Le fonctionnement sans clé est disponible dès le premier lancement.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Guide complet : [`integrations/hermes/`](../integrations/hermes/)
Démarrez le serveur de mémoire : `npx -y @agentmemory/agentmemory@latest`
-#### Skills natifs via `npx skills add` (50+ agents)
+#### Skills natifs via `npx skills add`
agentmemory livre 17 skills au format `<dir>/SKILL.md` façon Claude Code : 9 skills d'action invocables (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) et 8 skills de référence que l'agent charge à la demande (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Les skills de référence embarquent des tableaux de données générés depuis les sources, donc ils ne dérivent jamais. La CLI [`skills`](https://npmjs.com/package/skills) de vercel-labs les installe automatiquement dans le répertoire de skills natif de l'agent appelant sur 50+ agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, et plus) :
@@ -836,7 +628,7 @@ L'entrée agentmemory est le **même bloc de serveur MCP** sur tous les hôtes q
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installe MCP et les hooks de capture dans le répertoire de personnalisation partagé. Voir [Configuration et limites d'Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` utilise la même configuration MCP et hooks que les versions actuelles de l'IDE. Les installations existantes doivent se rafraîchir avec `--force` ; voir les [notes de mise à jour](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` écrit la config au niveau utilisateur. Les surcharges au niveau du workspace vont dans `.kiro/settings/mcp.json` à côté de votre code. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` écrit le bloc `mcpServers` standard. Warp découvre aussi automatiquement les skills depuis `.claude/skills/` ; une fois le plugin Claude Code installé, les 8 skills agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) apparaissent nativement dans la palette de commandes slash de Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` écrit le bloc `mcpServers` standard. Warp découvre aussi automatiquement les skills depuis `.claude/skills/` ; une fois le plugin Claude Code installé, les 9 skills agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) apparaissent nativement dans la palette de commandes slash de Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` écrit le bloc `mcpServers` standard. Utilisateurs de l'extension VS Code : collez le même bloc via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (préféré) ou `config.json` (legacy) | `agentmemory connect continue` crée une `config.yaml` absente ou modifie la `config.json` existante. Si le YAML existe, l'adaptateur affiche le bloc à ajouter sous `mcpServers:` pour préserver commentaires et ancres. Continue attend un tableau, pas un objet. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` écrit sous `context_servers` (la clé de Zed, PAS `mcpServers`). Les serveurs MCP distants peuvent être câblés via `{"url": "..."}` à la place. |
@@ -934,7 +726,6 @@ Chaque agent de codage IA est livré avec une mémoire intégrée : Claude Code
|---|---|---|
| Entretien | Instructions maintenues manuellement | Historique des sessions et souvenirs indexés |
| Recherche | Charge tout en contexte | BM25 + vectoriel + graphe (top-K seulement) |
-| Coût en tokens | 22K+ à 240 observations | ~1 900 tokens (92% de moins) |
| Entre agents | Fichiers par agent | MCP + REST (tout agent) |
| Coordination | Aucune | Leases, signaux, actions, routines |
| Observabilité | Lire les fichiers manuellement | Visualiseur temps réel sur :3113 |
@@ -947,7 +738,7 @@ Chaque agent de codage IA est livré avec une mémoire intégrée : Claude Code
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1032,7 +823,7 @@ Fusionné avec la Reciprocal Rank Fusion (RRF, k=60) et diversifié par session
Quand un index vectoriel est peuplé, `mem::search` (derrière `memory_recall`) utilise le ranker hybride BM25 + vectoriel. Sans embeddings, il utilise BM25. `smart-search` peut en plus fusionner des correspondances de graphe structurel quand des données de graphe existent, y compris en mode sans clé. Le recall de lessons tourne sur un index BM25 en mémoire dédié plutôt que de scanner tout le corpus à chaque requête. Les versions de mémoire supersédées sont exclues de tout chemin de recall ; la chaîne de versions conserve leur historique.
-Les vecteurs survivent à un crash ou à un force-kill. L'index vectoriel est sauvegardé par lots au plus toutes les `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutes). Chaque vecteur ajouté ou retiré entre-temps est aussi écrit immédiatement dans un petit journal en attente dans le state store, et le prochain démarrage le rejoue sans appeler le fournisseur d'embedding. Chaque sauvegarde réussie vide le journal. Les documents qui n'ont toujours pas de vecteur après le replay sont réembarqués en arrière-plan par lots de `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) jusqu'à épuisement, et un backfill arrêté reprend au démarrage suivant. `/agentmemory/status` et le visualiseur affichent la taille du journal en attente et l'état du backfill. Les installations sans clé n'écrivent rien.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenise le grec, le cyrillique, l'hébreu, l'arabe et le latin accentué nativement. Pour les mémoires en chinois / japonais / coréen, installez les segmenteurs optionnels (`npm install @node-rs/jieba tiny-segmenter`) pour découper les suites CJK en tokens au niveau mot ; sans eux, agentmemory retombe en douceur sur une tokenisation en bloc complet et affiche un indice ponctuel sur stderr.
@@ -1048,9 +839,9 @@ L'installation npm normale inclut le runtime optionnel `@huggingface/transformer
| Fournisseur | Modèle | Coût | Notes |
|---|---|---|---|
-| **Local (opt-in recommandé)** | `all-MiniLM-L6-v2` | Gratuit | Sur l'appareil après le premier téléchargement de modèle, +8pp de recall par rapport à BM25 seul |
+| **Local (opt-in recommandé)** | `all-MiniLM-L6-v2` | Gratuit | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Palier gratuit | 100+ langues, 768/1536/3072 dims (MRL), entrée de 2048 tokens. Remplace `text-embedding-004` ([obsolète, arrêt le 14 janvier 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | 0.02 $/1M | Meilleure qualité |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Payant | Optimisé pour le code |
| Cohere | `embed-english-v3.0` | Essai gratuit | Usage général |
| OpenRouter | N'importe quel modèle | Variable | Proxy multi-modèle |
diff --git a/READMEs/README.he-IL.md b/READMEs/README.he-IL.md
index 4eb23cb2..e18522b8 100644
--- a/READMEs/README.he-IL.md
+++ b/READMEs/README.he-IL.md
@@ -68,8 +68,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% שיעור שליפה R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% פחות טוקנים" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 כלי MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hooks אוטומטיים" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 מסדי נתונים חיצוניים" height="38" /></picture>
@@ -335,223 +333,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="מבחני ביצועים" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### דיוק שליפה
-
-**coding-agent-life-v1** (קורפוס פנימי, שניתן לשכפול ב-sandbox)
-
-| מתאם | P@5 | R@5 | שיעור פגיעה ב-Top-5 | latency p50 |
-|---|---|---|---|---|
-| **agentmemory היברידי** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep (קו בסיס) | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-שיעור פגיעה של 100% ב-top-5 בתקרה המתמטית של **P@5** עבור הקורפוס הזה (0.240, ראו scorecard). המצב ההיברידי שולף כל session זהב (gold); ל-grep חסר 1 מתוך 2 פריטי gold בשאילתה הטמפורלית המשתרעת על כמה sessions. השיפור הוא ב-**recall + טמפורליות**, לא בדיוק מצרפי. ה-benchmark הזה קטן ומעוט gold; LongMemEval-S הגדול יותר שלהלן מבדיל טוב יותר. פירוט מלא לפי טיפוס + הערת תיקון: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 שאלות)
+תוצאות האחזור העדכניות, הגדרות המדדים ושלבי השחזור:
-| מערכת | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| חלופת BM25 בלבד | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### חיסכון בטוקנים
-
-| גישה | טוקנים/שנה | עלות/שנה |
-|---|---|---|
-| הדבקת הקשר מלא | 19.5M+ | בלתי אפשרי (חורג מהחלון) |
-| סיכום על ידי LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embeddings מקומיים | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> מודל embedding: `all-MiniLM-L6-v2` (מקומי, חינמי, בלי מפתח API). דוחות מלאים: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). השוואת מתחרים: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) המכסה את agentmemory לעומת mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**שכפול מקומי:** [`eval/README.md`](../eval/README.md), harness עם adapters נתיקים עבור LongMemEval `_s` (500 שאלות ציבוריות) + `coding-agent-life-v1` (קורפוס פנימי של 15 sessions). מתאמי Grep / וקטור / agentmemory מקבלים ניקוד אחד לצד השני, פלט NDJSON, וה-scorecards המתפרסמים נוחתים ב-[`docs/benchmarks/`](../docs/benchmarks/).
-
-**משתלב עם [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), ו-[Graphify](https://github.com/safishamsi/graphify).** אינדוקס code-graph, pipelines לבנייה מרובת-סוכנים, וגרפי ידע רחבים יותר על פני docs / PDF / תמונות / וידאו. agentmemory זוכר את העבודה; שלושת הפרויקטים האלה מדליקים את שאר שכבת ההקשר. מתכונים + טבלת ניתוב שאלות: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="לעומת מתחרים" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Built-in (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>סוג</strong></td>
-<td>מנוע זיכרון + שרת MCP</td>
-<td>API של שכבת זיכרון</td>
-<td>runtime מלא לסוכן</td>
-<td>AI אישי</td>
-<td>API זיכרון + אפליקציה</td>
-<td>hub זיכרון לצוות (proxy ל-LLM)</td>
-<td>זיכרון וקטורי (קוד פתוח)</td>
-<td>מנוע זיכרון (Oracle DB)</td>
-<td>מערכת זיכרון</td>
-<td>קובץ סטטי</td>
-</tr>
-<tr>
-<td><strong>שליפה R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>דיווח עצמי</td>
-<td>PersonaMem 76% (דיווח עצמי)</td>
-<td>~96.6% (דיווח עצמי)</td>
-<td>94.4% (דיווח עצמי)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>לכידה אוטומטית</strong></td>
-<td>12 hooks (ללא מאמץ ידני)</td>
-<td>קריאות <code>add()</code> ידניות</td>
-<td>עריכות עצמיות של הסוכן</td>
-<td>ידני</td>
-<td>חילוץ בצד ה-API</td>
-<td>יירוט proxy (החלפת base-URL)</td>
-<td>ידני</td>
-<td>חילוץ API</td>
-<td>ידני</td>
-<td>עריכה ידנית</td>
-</tr>
-<tr>
-<td><strong>חיפוש</strong></td>
-<td>BM25 + וקטור + גרף (fusion מסוג RRF)</td>
-<td>וקטור + גרף</td>
-<td>וקטור (ארכיוני)</td>
-<td>סמנטי</td>
-<td>וקטור + RAG</td>
-<td>4 סוגי נכסים (Chat / Skill / Wiki / CodeGraph)</td>
-<td>וקטור בלבד</td>
-<td>וקטור + סמנטי</td>
-<td>משוקלל decay</td>
-<td>טוען הכול להקשר</td>
-</tr>
-<tr>
-<td><strong>ריבוי-סוכנים</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (בלי תיאום)</td>
-<td>רק בתוך runtime של Letta</td>
-<td>לא</td>
-<td>לא</td>
-<td>תפקידי צוות + נכסים משותפים</td>
-<td>לא</td>
-<td>מוגבל לתחום בלבד</td>
-<td>משותף בין סוכנים מרובים</td>
-<td>קבצים לפי סוכן</td>
-</tr>
-<tr>
-<td><strong>נעילה ל-framework</strong></td>
-<td>אין (כל לקוח MCP)</td>
-<td>אין</td>
-<td>גבוהה (חייבים להשתמש ב-Letta)</td>
-<td>עצמאי (standalone)</td>
-<td>אין</td>
-<td>Proxy חוצץ בפני כל קריאת מודל</td>
-<td>אין</td>
-<td>Oracle Database</td>
-<td>אין</td>
-<td>פורמט לפי סוכן</td>
-</tr>
-<tr>
-<td><strong>תלויות חיצוניות</strong></td>
-<td>אין צורך בשירות מסד נתונים נפרד</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + מסד נתונים וקטורי</td>
-<td>מרובות</td>
-<td>cloud מנוהל</td>
-<td>ערימת Docker (Core + Hub + Proxy)</td>
-<td>מאגר וקטורי</td>
-<td>Oracle AI Database</td>
-<td>אין</td>
-<td>אין</td>
-</tr>
-<tr>
-<td><strong>מחזור חיים של הזיכרון</strong></td>
-<td>קונסולידציה ב-4 שכבות + decay + שכחה אוטומטית</td>
-<td>חילוץ פסיבי</td>
-<td>מנוהל על ידי הסוכן</td>
-<td>ידני</td>
-<td>שכחה אוטומטית</td>
-<td>בדיקה ידנית; ניתוב אוטומטי בפיתוח</td>
-<td>אין</td>
-<td>לא מצוין</td>
-<td>decay + קונסולידציה</td>
-<td>גיזום ידני</td>
-</tr>
-<tr>
-<td><strong>יעילות טוקנים</strong></td>
-<td>~1,900 טוקנים/session ($10/שנה)</td>
-<td>משתנה לפי אינטגרציה</td>
-<td>זיכרון core בתוך ההקשר</td>
-<td>משתנה</td>
-<td>תמחור cloud</td>
-<td>לא מצוין</td>
-<td>אין תקציב טוקנים</td>
-<td>מגובה LLM (משתנה)</td>
-<td>משתנה</td>
-<td>22K+ טוקנים ב-240 תצפיות</td>
-</tr>
-<tr>
-<td><strong>מציג בזמן אמת</strong></td>
-<td>כן (פורט 3113)</td>
-<td>dashboard ב-cloud</td>
-<td>dashboard ב-cloud</td>
-<td>UI מבוסס web</td>
-<td>dashboard ב-cloud</td>
-<td>UI web של ה-hub</td>
-<td>לא</td>
-<td>לא</td>
-<td>לא</td>
-<td>לא</td>
-</tr>
-<tr>
-<td><strong>אירוח עצמי</strong></td>
-<td>כן (ברירת מחדל)</td>
-<td>אופציונלי</td>
-<td>אופציונלי</td>
-<td>כן</td>
-<td>לא (cloud בלבד)</td>
-<td>כן (Docker)</td>
-<td>כן</td>
-<td>כן (Oracle DB)</td>
-<td>כן</td>
-<td>כן</td>
-</tr>
-</table>
-
-<sub>הערת benchmark: רק ה-R@5 של agentmemory הוא תוצאת מדידה שלנו (LongMemEval-S, ניתנת לשכפול מ-<a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). המספרים של mem0 ו-Letta הם המספרים המתפרסמים שלהם מ-LoCoMo (dataset שונה); המספרים של MemPalace, supermemory, TencentDB (PersonaMem), ו-oracleagentmemory הם טענות שדווחו על ידי היצרן ולא שכפלנו אותן באופן עצמאי (ההרצה של oracleagentmemory השתמשה ב-GPT-5.5 מול Oracle AI Database). מוצגים זה לצד זה להערכה כללית בלבד, לא כהשוואת ראש-בראש על אותם נתונים. ספירות הכוכבים הן משוערות ומשתנות עם הזמן.</sub>
-
-**שחקנים חדשים יותר** שכדאי להכיר, מושווים בהרחבה ב-[`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| מערכת | ⭐ | זווית |
-|--------|---|-------|
-| Zep / Graphiti | 30K | גרף ידע טמפורלי; תוצאות השאילתה הטמפורלית הפרסומיות החזקות ביותר (LongMemEval 63.8%), אבל הגרף נבנה באופן אסינכרוני כך שעובדות חדשות יכולות להתעכב |
-| Cognee | 30K | בליעת מסמכים לגרף ידע (document-to-knowledge-graph), Python בלבד, נבנה לחילוץ ישויות מבני ולא ללכידת sessions |
+השוואה מעודכנת של מוצרים ואפשרויות פריסה, עם מקורות:
-AgentMemory משלב תיעוד סשנים, היסטוריית פרויקט שניתן לחפש בה וממשק צפייה מקומי, עם עבודה ללא מפתח כבר מההפעלה הראשונה.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -788,7 +580,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
הפעילו את שרת הזיכרון: `npx -y @agentmemory/agentmemory@latest`
-#### Skills מובנים דרך `npx skills add` (50+ סוכנים)
+#### Skills מובנים דרך `npx skills add`
agentmemory שולח 17 skills בפורמט `<dir>/SKILL.md` בסטייל Claude Code: 9 skills של פעולה שניתן להפעיל (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) ו-8 skills של עיון שהסוכן טוען לפי הצורך (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). ה-skills של העיון נושאים טבלאות נתונים שנוצרות מהמקור, כך שהן לעולם לא נסחפות (drift). ה-CLI [`skills`](https://npmjs.com/package/skills) של vercel-labs מתקין אותם אוטומטית לתוך תיקיית ה-skills המובנית של הסוכן הקורא על פני 50+ סוכנים (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, ועוד):
@@ -846,7 +638,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` מתקין MCP ו-hooks של לכידה בתיקיית ההתאמה האישית המשותפת. ראו [הגדרה ומגבלות של Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` משתמש באותה תצורת MCP ו-hooks כמו גרסאות ה-IDE הנוכחיות. התקנות קיימות צריכות לרפרש עם `--force`; ראו את [הערות השדרוג](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` כותב את התצורה בשכבת המשתמש. override-ים בשכבת ה-workspace נכנסים ב-`.kiro/settings/mcp.json` לצד הקוד שלכם. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` כותב את בלוק ה-`mcpServers` הסטנדרטי. Warp גם מגלה אוטומטית skills מ-`.claude/skills/`; כשה-plugin של Claude Code מותקן, 8 ה-skills של agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) מופיעים באופן מובנה בפלטת פקודות ה-slash של Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` כותב את בלוק ה-`mcpServers` הסטנדרטי. Warp גם מגלה אוטומטית skills מ-`.claude/skills/`; כשה-plugin של Claude Code מותקן, 9 ה-skills של agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) מופיעים באופן מובנה בפלטת פקודות ה-slash של Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` כותב את בלוק ה-`mcpServers` הסטנדרטי. למשתמשי ה-extension של VS Code: הדביקו את אותו בלוק דרך Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (מועדף) או `config.json` (legacy) | `agentmemory connect continue` יוצר `config.yaml` כשאין הגדרות, או מעדכן `config.json` קיים. עבור YAML קיים הוא מציג בלוק להוספה תחת `mcpServers:`, כדי לשמור הערות ועוגנים. Continue משתמש במערך ולא באובייקט. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` כותב תחת `context_servers` (המפתח של Zed, לא `mcpServers`). שרתי MCP מרוחקים אפשר לחווט במקום זאת דרך `{"url": "..."}`. |
@@ -928,7 +720,6 @@ REST משתמש ב-`3111` וממשק הצפייה ב-`3113`. השאירו אות
|---|---|---|
| תחזוקה | הוראות שנערכות ידנית | היסטוריית סשנים וזיכרונות עם אינדקס |
| חיפוש | טוען הכול להקשר | BM25 + וקטור + גרף (top-K בלבד) |
-| עלות טוקנים | 22K+ ב-240 תצפיות | ~1,900 טוקנים (92% פחות) |
| בין-סוכנים | קבצים לפי סוכן | MCP + REST (כל סוכן) |
| תיאום | אין | leases, signals, actions, routines |
| Observability | קריאת קבצים ידנית | מציג בזמן אמת בפורט :3113 |
@@ -941,7 +732,7 @@ REST משתמש ב-`3111` וממשק הצפייה ב-`3113`. השאירו אות
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1026,7 +817,7 @@ SessionStart hook fires
כשאינדקס וקטורי מאוכלס, `mem::search` (מאחורי `memory_recall`) משתמש ב-ranker ההיברידי של BM25 + וקטור. בלי embeddings הוא משתמש ב-BM25. `smart-search` יכול גם למזג התאמות מבניות מהגרף כשנתוני גרף קיימים, גם במצב keyless. שליפת lessons רצה על אינדקס BM25 ייעודי בזיכרון במקום לסקור את כל הקורפוס בכל שאילתה. גרסאות זיכרון שהוחלפו (superseded) מוחרגות מכל נתיב שליפה; שרשרת הגרסאות שומרת את ההיסטוריה שלהן.
-וקטורים שורדים קריסה או force-kill. אינדקס הווקטורים נשמר ב-buckets לכל היותר כל `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 דקות). כל וקטור שנוסף או הוסר בינתיים גם נכתב מיידית ל-log קטן של pending ב-state store, וההפעלה הבאה משדרת אותו (replay) מחדש בלי לקרוא לספק ה-embedding. כל שמירה מוצלחת מרוקנת את ה-log. מסמכים שעדיין אין להם וקטור אחרי ה-replay מוטבעים (re-embedded) מחדש ברקע ב-batches של `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) עד שלא נשאר אף אחד, ו-backfill שנעצר ממשיך בהפעלה הבאה. `/agentmemory/status` והמציג מציגים את גודל ה-log הממתין ואת מצב ה-backfill. התקנות keyless לא כותבות שום דבר.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 מבצע tokenization ליוונית, קירילית, עברית, ערבית, ולטינית עם סימני הטעמה ישר מהקופסה. לזיכרונות בסינית / יפנית / קוריאנית, התקינו את ה-segmenters האופציונליים (`npm install @node-rs/jieba tiny-segmenter`) כדי לפצל רצפי CJK ל-tokens בגובה-מילה; בלעדיהם, agentmemory נופל בעדינות ל-tokenization של הרצף השלם ומדפיס רמז חד-פעמי ל-stderr.
@@ -1042,9 +833,9 @@ EMBEDDING_PROVIDER=local
| ספק | מודל | עלות | הערות |
|---|---|---|---|
-| **מקומי (opt-in מומלץ)** | `all-MiniLM-L6-v2` | חינמי | במכשיר עצמו אחרי הורדת המודל הראשונה, +8pp ל-recall לעומת BM25 בלבד |
+| **מקומי (opt-in מומלץ)** | `all-MiniLM-L6-v2` | חינמי | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | שכבה חינמית | 100+ שפות, 768/1536/3072 מימדים (MRL), קלט של 2048 טוקנים. מחליף את `text-embedding-004` ([הוצא משימוש, כיבוי ב-14 בינואר 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | האיכות הגבוהה ביותר |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | בתשלום | מותאם לקוד |
| Cohere | `embed-english-v3.0` | ניסיון חינמי | כללי |
| OpenRouter | כל מודל | משתנה | proxy מרובה-מודלים |
diff --git a/READMEs/README.hi-IN.md b/READMEs/README.hi-IN.md
index 5963fa93..efee2e6e 100644
--- a/READMEs/README.hi-IN.md
+++ b/READMEs/README.hi-IN.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -331,223 +329,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval सटीकता
-
-**coding-agent-life-v1** (in-house corpus, sandbox-reproducible)
-
-| Adapter | P@5 | R@5 | Top-5 hit rate | p50 latency |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-इस corpus के लिए **P@5 math ceiling** (0.240, scorecard देखें) पर 100% top-5 hit rate। Hybrid हर gold session retrieve करता है; grep multi-session temporal query पर 2 में से 1 gold miss करता है। Lift **recall + temporal** है, aggregate precision नहीं। यह benchmark छोटा और gold-sparse है; नीचे का बड़ा LongMemEval-S बेहतर differentiate करता है। पूरी per-type breakdown + correction नोट: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)।
-
-**LongMemEval-S** (ICLR 2025, 500 questions)
+वर्तमान पुनर्प्राप्ति परिणाम, मेट्रिक की परिभाषाएँ और परीक्षण दोहराने के चरण:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token बचत
-
-| दृष्टिकोण | Tokens/yr | Cost/yr |
-|---|---|---|
-| पूरा context paste करें | 19.5M+ | असंभव (window से अधिक) |
-| LLM-summarized | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + local embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding model: `all-MiniLM-L6-v2` (local, free, कोई API key नहीं)। पूरी रिपोर्ट्स: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md)। प्रतिस्पर्धी तुलना: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), जो agentmemory बनाम mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo को cover करती है।
-
-**स्थानीय रूप से reproduce करें:** [`eval/README.md`](../eval/README.md), LongMemEval `_s` (public 500-Q) + `coding-agent-life-v1` (in-house 15-session corpus) के लिए एक adapter-pluggable harness। Grep / vector / agentmemory adapters साथ-साथ score होते हैं, NDJSON output, published scorecards [`docs/benchmarks/`](../docs/benchmarks/) में जाते हैं।
-
-**[codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), और [Graphify](https://github.com/safishamsi/graphify) के साथ जोड़ता है।** Code-graph indexing, multi-agent build pipelines, और docs / PDFs / images / videos में व्यापक knowledge graphs। agentmemory काम याद रखता है; ये तीन projects context layer के बाकी हिस्से को रोशन करते हैं। Recipes + question-routing table: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md)।
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>बिल्ट-इन (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>प्रकार</strong></td>
-<td>Memory engine + MCP सर्वर</td>
-<td>Memory layer API</td>
-<td>पूर्ण agent runtime</td>
-<td>Personal AI</td>
-<td>Memory API + app</td>
-<td>Team memory hub (LLM proxy)</td>
-<td>Vector memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Memory system</td>
-<td>Static फाइल</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Self-reported</td>
-<td>PersonaMem 76% (self-reported)</td>
-<td>~96.6% (self-reported)</td>
-<td>94.4% (self-reported)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>स्वचालित capture</strong></td>
-<td>12 hooks (शून्य मैनुअल प्रयास)</td>
-<td>मैनुअल <code>add()</code> कॉल</td>
-<td>एजेंट self-edits</td>
-<td>मैनुअल</td>
-<td>API-side extraction</td>
-<td>Proxy interception (base-URL swap)</td>
-<td>मैनुअल</td>
-<td>API extraction</td>
-<td>मैनुअल</td>
-<td>मैनुअल editing</td>
-</tr>
-<tr>
-<td><strong>खोज</strong></td>
-<td>BM25 + Vector + Graph (RRF fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semantic</td>
-<td>Vector + RAG</td>
-<td>4 asset types (Chat / Skill / Wiki / CodeGraph)</td>
-<td>केवल-vector</td>
-<td>Vector + semantic</td>
-<td>Decay-weighted</td>
-<td>सब कुछ context में load करता है</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (कोई coordination नहीं)</td>
-<td>केवल Letta runtime में</td>
-<td>नहीं</td>
-<td>नहीं</td>
-<td>Team roles + shared assets</td>
-<td>नहीं</td>
-<td>केवल scoped</td>
-<td>Multi-agent shared</td>
-<td>प्रति-एजेंट फाइलें</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>कोई नहीं (कोई भी MCP क्लाइंट)</td>
-<td>कोई नहीं</td>
-<td>उच्च (Letta का उपयोग आवश्यक)</td>
-<td>Standalone</td>
-<td>कोई नहीं</td>
-<td>Proxy हर model call के सामने रहता है</td>
-<td>कोई नहीं</td>
-<td>Oracle Database</td>
-<td>कोई नहीं</td>
-<td>प्रति-एजेंट format</td>
-</tr>
-<tr>
-<td><strong>बाहरी निर्भरताएँ</strong></td>
-<td>अलग डेटाबेस सेवा की जरूरत नहीं</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>कई</td>
-<td>Managed cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>कोई नहीं</td>
-<td>कोई नहीं</td>
-</tr>
-<tr>
-<td><strong>Memory lifecycle</strong></td>
-<td>4-tier consolidation + decay + auto-forget</td>
-<td>Passive extraction</td>
-<td>Agent-managed</td>
-<td>मैनुअल</td>
-<td>Auto-forget</td>
-<td>मैनुअल review; auto-routing प्रगति पर</td>
-<td>कोई नहीं</td>
-<td>बताया नहीं गया</td>
-<td>Decay + consolidation</td>
-<td>मैनुअल pruning</td>
-</tr>
-<tr>
-<td><strong>Token दक्षता</strong></td>
-<td>~1,900 tokens/session ($10/yr)</td>
-<td>integration पर निर्भर</td>
-<td>Core memory context में</td>
-<td>भिन्न</td>
-<td>Cloud pricing</td>
-<td>बताया नहीं गया</td>
-<td>कोई token budget नहीं</td>
-<td>LLM-backed (भिन्न)</td>
-<td>भिन्न</td>
-<td>240 observations पर 22K+ tokens</td>
-</tr>
-<tr>
-<td><strong>Real-time व्यूअर</strong></td>
-<td>हाँ (port 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>नहीं</td>
-<td>नहीं</td>
-<td>नहीं</td>
-<td>नहीं</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>हाँ (default)</td>
-<td>Optional</td>
-<td>Optional</td>
-<td>हाँ</td>
-<td>नहीं (केवल-cloud)</td>
-<td>हाँ (Docker)</td>
-<td>हाँ</td>
-<td>हाँ (Oracle DB)</td>
-<td>हाँ</td>
-<td>हाँ</td>
-</tr>
-</table>
-
-<sub>Benchmark नोट: केवल agentmemory का R@5 हमारा अपना measured result है (LongMemEval-S, <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> से reproducible)। mem0 और Letta के आँकड़े उनके published LoCoMo numbers हैं (एक अलग dataset); MemPalace, supermemory, TencentDB (PersonaMem), और oracleagentmemory के आँकड़े vendor self-reported दावे हैं जिन्हें हमने स्वतंत्र रूप से reproduce नहीं किया है (oracleagentmemory की run ने Oracle AI Database के विरुद्ध GPT-5.5 का उपयोग किया)। केवल ballpark के लिए साथ-साथ दिखाए गए हैं, समान data पर head-to-head तुलना नहीं। Star counts अनुमानित हैं और समय के साथ बदलते रहते हैं।</sub>
-
-**नए प्रवेशक** जिन्हें जानना उपयोगी है, [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) में गहराई से compare किए गए:
-
-| System | ⭐ | Angle |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal knowledge graph; सबसे मज़बूत published temporal-query results (LongMemEval 63.8%), लेकिन graph asynchronously build होता है इसलिए ताज़ा facts पिछड़ सकते हैं |
-| Cognee | 30K | Document-to-knowledge-graph ingestion, केवल-Python, session capture के बजाय structured entity extraction के लिए बना |
+स्रोतों सहित उत्पादों और परिनियोजन विकल्पों की अद्यतन तुलना:
-AgentMemory सत्र capture, खोजने योग्य प्रोजेक्ट इतिहास और लोकल व्यूअर साथ देता है; पहली बार से ही बिना कुंजी के काम कर सकते हैं।
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
memory server शुरू करें: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add` के माध्यम से Native skills (50+ agents)
+#### `npx skills add` के माध्यम से Native skills
agentmemory Claude-Code-style `<dir>/SKILL.md` format में 17 skills ship करता है: 9 invocable action skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) और 8 reference skills जिन्हें agent on demand load करता है (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`)। Reference skills में source से generated data tables होती हैं, इसलिए वे कभी drift नहीं करती। vercel-labs की [`skills`](https://npmjs.com/package/skills) CLI इन्हें calling agent की native skill directory में 50+ agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, और अधिक) में auto-install करती है:
@@ -837,7 +629,7 @@ agentmemory entry, `mcpServers` shape का उपयोग करने वा
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` shared customization directory में MCP और capture hooks install करता है। देखें [Antigravity setup और limits](../docs/plugins/antigravity.md)। |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` current IDE versions जैसा ही MCP और hook configuration उपयोग करता है। मौजूदा installations को `--force` से refresh करना चाहिए; देखें [upgrade notes](../docs/plugins/antigravity.md)। |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` user-level config लिखता है। Workspace overrides आपके code के बगल में `.kiro/settings/mcp.json` में जाते हैं। |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` standard `mcpServers` block लिखता है। Warp `.claude/skills/` से skills भी auto-discover करता है; Claude Code plugin install होने के बाद 8 agentmemory skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) Warp की slash-command palette में natively दिखती हैं। |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` standard `mcpServers` block लिखता है। Warp `.claude/skills/` से skills भी auto-discover करता है; Claude Code plugin install होने के बाद 9 agentmemory skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) Warp की slash-command palette में natively दिखती हैं। |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` standard `mcpServers` block लिखता है। VS Code extension users: वही block Cline Settings → MCP Servers → Edit JSON के माध्यम से paste करें। |
| **Continue.dev** | `~/.continue/config.yaml` (preferred) या `config.json` (legacy) | `agentmemory connect continue` कॉन्फ़िगरेशन न होने पर `config.yaml` बनाता है या मौजूद `config.json` अपडेट करता है। पहले से मौजूद YAML के comments और anchors बचाने के लिए उसे नहीं बदलता, बल्कि `mcpServers:` के नीचे पेस्ट करने का block दिखाता है। Continue का `mcpServers` एक array है। |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` `context_servers` (Zed की key, `mcpServers` नहीं) के अंतर्गत लिखता है। Remote MCP servers इसके बजाय `{"url": "..."}` के माध्यम से wired हो सकते हैं। |
@@ -936,7 +728,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| देखभाल | हाथ से व्यवस्थित निर्देश | इंडेक्स किया गया सत्र इतिहास और मेमोरी |
| खोज | सब कुछ context में load करता है | BM25 + vector + graph (केवल top-K) |
-| Token cost | 240 observations पर 22K+ | ~1,900 tokens (92% कम) |
| Cross-agent | प्रति-agent फाइलें | MCP + REST (कोई भी agent) |
| Coordination | कोई नहीं | Leases, signals, actions, routines |
| Observability | फाइलें मैनुअल पढ़ें | :3113 पर real-time viewer |
@@ -949,7 +740,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1034,7 +825,7 @@ Reciprocal Rank Fusion (RRF, k=60) के साथ fuse होता है औ
जब vector index populated होता है, तो `mem::search` (`memory_recall` के पीछे) hybrid BM25 + vector ranker का उपयोग करता है। Embeddings के बिना यह BM25 का उपयोग करता है। Graph data मौजूद होने पर `smart-search` अतिरिक्त रूप से structural graph matches को भी fuse कर सकता है, keyless mode में भी। Lesson recall हर query पर पूरे corpus को scan करने के बजाय एक dedicated in-memory BM25 index पर चलती है। Superseded memory versions हर recall path से excluded हैं; version chain उनका history रखती है।
-Vectors किसी crash या force-kill से बच जाते हैं। Vector index को कम से कम हर `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 मिनट) पर buckets में save किया जाता है। इसके बीच जोड़ा या हटाया गया हर vector भी state store में एक छोटे pending log में तुरंत लिखा जाता है, और अगला start embedding provider को call किए बिना उसे replay करता है। हर successful save उस log को खाली कर देता है। जिन documents के पास replay के बाद भी कोई vector नहीं है उन्हें background में `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) के batches में तब तक re-embed किया जाता है जब तक कोई बाकी न रहे, और रुका हुआ backfill अगले start पर जारी रहता है। `/agentmemory/status` और viewer pending log का size और backfill state दिखाते हैं। Keyless installs कुछ नहीं लिखते।
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 box से बाहर ही Greek, Cyrillic, Hebrew, Arabic, और accented Latin को tokenize करता है। Chinese / Japanese / Korean memories के लिए, CJK runs को word-level tokens में split करने के लिए optional segmenters install करें (`npm install @node-rs/jieba tiny-segmenter`); उनके बिना, agentmemory soft-fall back होकर whole-run tokenization पर जाता है और stderr पर एक-बार hint print करता है।
@@ -1050,9 +841,9 @@ Normal npm install में optional `@huggingface/transformers` runtime शा
| Provider | Model | Cost | नोट्स |
|---|---|---|---|
-| **Local (अनुशंसित opt-in)** | `all-MiniLM-L6-v2` | Free | पहले model download के बाद on-device, BM25-only पर +8pp recall |
+| **Local (अनुशंसित opt-in)** | `all-MiniLM-L6-v2` | Free | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free tier | 100+ भाषाएँ, 768/1536/3072 dims (MRL), 2048-token input। `text-embedding-004` को replace करता है ([deprecated, 14 जनवरी 2026 को shutdown](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | उच्चतम quality |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Paid | Code के लिए optimized |
| Cohere | `embed-english-v3.0` | Free trial | General purpose |
| OpenRouter | कोई भी model | भिन्न | Multi-model proxy |
diff --git a/READMEs/README.hu-HU.md b/READMEs/README.hu-HU.md
index fc2ae766..a5e23b1e 100644
--- a/READMEs/README.hu-HU.md
+++ b/READMEs/README.hu-HU.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% visszakeresési R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92%-kal kevesebb token" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP eszköz" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automatikus hook" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 külső adatbázis" height="38" /></picture>
@@ -331,223 +329,17 @@ A legfrissebb kiadási megjegyzések: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarkok" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Visszakeresési pontosság
-
-**coding-agent-life-v1** (házon belüli korpusz, sandboxban reprodukálható)
-
-| Adapter | P@5 | R@5 | Top-5 találati arány | p50 késleltetés |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep alapvonal | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100%-os top-5 találati arány a **P@5 matematikai felső határán** ennél a korpusznál (0.240, lásd a scorecardot). A hibrid minden gold sessiont visszakeres; a grep 2-ből 1 gold eredményt elveszít a több sessiont átfogó, időbeli lekérdezésnél. A nyereség **felidézésben és időbeliségben** mutatkozik, nem az összesített pontosságban. Ez a benchmark kicsi és gold-szegény; a lejjebb található, nagyobb LongMemEval-S jobban differenciál. Teljes, típus szerinti bontás + korrekciós megjegyzés: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 kérdés)
+Aktuális visszakeresési eredmények, mérőszámok és reprodukálási lépések:
-| Rendszer | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Csak BM25 fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token-megtakarítás
-
-| Megközelítés | Token/év | Költség/év |
-|---|---|---|
-| Teljes kontextus beillesztése | 19.5M+ | Lehetetlen (meghaladja az ablakot) |
-| LLM által összefoglalva | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokális embeddingek | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding modell: `all-MiniLM-L6-v2` (lokális, ingyenes, nincs szükség API-kulcsra). Teljes jelentések: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Versenytárs-összehasonlítás: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), amely az agentmemoryt a mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo rendszerekkel veti össze.
-
-**Lokális reprodukálás:** [`eval/README.md`](../eval/README.md), egy adapter-bővíthető harness a LongMemEval `_s` (nyilvános, 500 kérdéses) + `coding-agent-life-v1` (házon belüli, 15 sessiont tartalmazó korpusz) számára. A grep / vektor / agentmemory adapterek egymás mellett kapnak pontszámot, NDJSON kimenettel, a publikált scorecardok a [`docs/benchmarks/`](../docs/benchmarks/) mappában landolnak.
-
-**Jól illik a [codegraph](https://github.com/colbymchenry/codegraph), az [Understand Anything](https://github.com/Lum1104/Understand-Anything) és a [Graphify](https://github.com/safishamsi/graphify) mellé.** Kódgráf-indexelés, több ágenses build-pipeline-ok, és szélesebb tudásgráfok dokumentumok / PDF-ek / képek / videók fölött. Az agentmemory a munkára emlékszik; ez a három projekt a kontextusréteg többi részét kelti életre. Receptek + kérdés-útválasztási táblázat: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Versenytársakkal szemben" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Beépített (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Típus</strong></td>
-<td>Memóriamotor + MCP szerver</td>
-<td>Memóriaréteg API</td>
-<td>Teljes ágens-futtatókörnyezet</td>
-<td>Személyes AI</td>
-<td>Memória API + app</td>
-<td>Csapatmemória-hub (LLM proxy)</td>
-<td>Vektormemória (OSS)</td>
-<td>Memóriamotor (Oracle DB)</td>
-<td>Memóriarendszer</td>
-<td>Statikus fájl</td>
-</tr>
-<tr>
-<td><strong>Visszakeresési R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Önbevallás alapján</td>
-<td>PersonaMem 76% (önbevallás alapján)</td>
-<td>~96.6% (önbevallás alapján)</td>
-<td>94.4% (önbevallás alapján)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Automatikus rögzítés</strong></td>
-<td>12 hook (nulla manuális munka)</td>
-<td>Manuális <code>add()</code> hívások</td>
-<td>Az ágens saját szerkesztései</td>
-<td>Manuális</td>
-<td>API-oldali kinyerés</td>
-<td>Proxy-elfogás (base-URL csere)</td>
-<td>Manuális</td>
-<td>API kinyerés</td>
-<td>Manuális</td>
-<td>Manuális szerkesztés</td>
-</tr>
-<tr>
-<td><strong>Keresés</strong></td>
-<td>BM25 + Vektor + Gráf (RRF fúzió)</td>
-<td>Vektor + Gráf</td>
-<td>Vektor (archivális)</td>
-<td>Szemantikus</td>
-<td>Vektor + RAG</td>
-<td>4 eszköztípus (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Csak vektor</td>
-<td>Vektor + szemantikus</td>
-<td>Lecsengés-súlyozott</td>
-<td>Mindent betölt a kontextusba</td>
-</tr>
-<tr>
-<td><strong>Több ágens</strong></td>
-<td>MCP + REST + lease-ek + szignálok</td>
-<td>API (nincs koordináció)</td>
-<td>Csak a Letta futtatókörnyezeten belül</td>
-<td>Nem</td>
-<td>Nem</td>
-<td>Csapatszerepek + megosztott eszközök</td>
-<td>Nem</td>
-<td>Csak hatókörön belül</td>
-<td>Több ágens közt megosztva</td>
-<td>Ágensenkénti fájlok</td>
-</tr>
-<tr>
-<td><strong>Keretrendszer-függés</strong></td>
-<td>Nincs (bármely MCP kliens)</td>
-<td>Nincs</td>
-<td>Magas (a Lettát kell használni)</td>
-<td>Önálló</td>
-<td>Nincs</td>
-<td>A proxy minden modellhívás elé kerül</td>
-<td>Nincs</td>
-<td>Oracle Database</td>
-<td>Nincs</td>
-<td>Ágensenkénti formátum</td>
-</tr>
-<tr>
-<td><strong>Külső függőségek</strong></td>
-<td>Nem szükséges külön adatbázis-szolgáltatás</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektor DB</td>
-<td>Több</td>
-<td>Felügyelt felhő</td>
-<td>Docker-stack (Core + Hub + Proxy)</td>
-<td>Vektortár</td>
-<td>Oracle AI Database</td>
-<td>Nincs</td>
-<td>Nincs</td>
-</tr>
-<tr>
-<td><strong>Memória életciklus</strong></td>
-<td>4 szintű konszolidáció + lecsengés + automatikus elfelejtés</td>
-<td>Passzív kinyerés</td>
-<td>Az ágens kezeli</td>
-<td>Manuális</td>
-<td>Automatikus elfelejtés</td>
-<td>Manuális átnézés; az automatikus útválasztás folyamatban</td>
-<td>Nincs</td>
-<td>Nincs megadva</td>
-<td>Lecsengés + konszolidáció</td>
-<td>Manuális ritkítás</td>
-</tr>
-<tr>
-<td><strong>Token-hatékonyság</strong></td>
-<td>~1,900 token/session ($10/év)</td>
-<td>Az integrációtól függ</td>
-<td>A core memória a kontextusban van</td>
-<td>Változó</td>
-<td>Felhő alapú díjszabás</td>
-<td>Nincs megadva</td>
-<td>Nincs token-budget</td>
-<td>LLM-alapú (változó)</td>
-<td>Változó</td>
-<td>22K+ token 240 megfigyelésnél</td>
-</tr>
-<tr>
-<td><strong>Valós idejű megjelenítő</strong></td>
-<td>Igen (3113-as port)</td>
-<td>Felhő alapú dashboard</td>
-<td>Felhő alapú dashboard</td>
-<td>Webes felhasználói felület</td>
-<td>Felhő alapú dashboard</td>
-<td>Hub webes felhasználói felület</td>
-<td>Nincs</td>
-<td>Nincs</td>
-<td>Nincs</td>
-<td>Nincs</td>
-</tr>
-<tr>
-<td><strong>Önállóan üzemeltethető</strong></td>
-<td>Igen (alapértelmezett)</td>
-<td>Opcionális</td>
-<td>Opcionális</td>
-<td>Igen</td>
-<td>Nem (csak felhő)</td>
-<td>Igen (Docker)</td>
-<td>Igen</td>
-<td>Igen (Oracle DB)</td>
-<td>Igen</td>
-<td>Igen</td>
-</tr>
-</table>
-
-<sub>Megjegyzés a benchmarkhoz: csak az agentmemory R@5 értéke a saját, mért eredményünk (LongMemEval-S, reprodukálható a <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> alapján). A mem0 és a Letta számai a saját publikált LoCoMo-adataik (egy másik adathalmaz); a MemPalace, a supermemory, a TencentDB (PersonaMem) és az oracleagentmemory számai a gyártók önbevallás alapú állításai, amelyeket nem reprodukáltunk függetlenül (az oracleagentmemory futása GPT-5.5-öt használt egy Oracle AI Database ellen). Csak tájékoztató jellegű egymás mellett közlésre szánva, nem azonos adatokon futó közvetlen összehasonlításra. A csillagszámok közelítők, és idővel változnak.</sub>
-
-**Újabb belépők**, amelyeket érdemes ismerni, részletesen összehasonlítva a [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) fájlban:
-
-| Rendszer | ⭐ | Megközelítés |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporális tudásgráf; a legerősebb publikált időbeli lekérdezési eredmények (LongMemEval 63.8%), de a gráf aszinkron módon épül, így a friss tények lemaradhatnak |
-| Cognee | 30K | Dokumentum-tudásgráf feldolgozás, csak Python, strukturált entitáskinyerésre épül, nem session-rögzítésre |
+A termékek és telepítési lehetőségek frissített összehasonlítása, forrásokkal:
-Az AgentMemory egyesíti a munkamenetek rögzítését, a kereshető projektelőzményeket és a helyi nézőfelületet, az első indítástól kulcs nélküli működéssel.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -784,7 +576,7 @@ Teljes útmutató: [`integrations/hermes/`](../integrations/hermes/)
Indítsd el a memóriaszervert: `npx -y @agentmemory/agentmemory@latest`
-#### Natív skillek az `npx skills add` révén (50+ ágens)
+#### Natív skillek az `npx skills add` révén
Az agentmemory 17 skillt szállít a Claude-Code-stílusú `<dir>/SKILL.md` formátumban: 9 meghívható akció-skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) és 8 referencia-skill, amelyeket az ágens igény szerint tölt be (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). A referencia-skillek forrásból generált adattáblákat hordoznak, így soha nem térnek el. A [`skills`](https://npmjs.com/package/skills) CLI, a vercel-labs eszköze, automatikusan telepíti ezeket a hívó ágens natív skill-mappájába, 50+ ágens között (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, és még több):
@@ -842,7 +634,7 @@ Az agentmemory bejegyzés **ugyanaz az MCP szerver blokk** minden olyan hoszton,
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | Az `agentmemory connect antigravity --with-hooks` telepíti az MCP-t és a rögzítő hookokat a megosztott testreszabási mappába. Lásd az [Antigravity beállítása és korlátai](../docs/plugins/antigravity.md) útmutatót. |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | Az `agentmemory connect antigravity-cli --with-hooks` ugyanazt az MCP- és hook-konfigurációt használja, mint a jelenlegi IDE-verziók. A meglévő telepítéseknek frissülniük kell a `--force`-szal; lásd a [frissítési megjegyzéseket](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | Az `agentmemory connect kiro` beírja a felhasználói szintű konfigurációt. A workspace-szintű felülírások a `.kiro/settings/mcp.json`-ba kerülnek, a kódod mellé. |
-| **Warp** | `~/.warp/.mcp.json` | Az `agentmemory connect warp` beírja a standard `mcpServers` blokkot. A Warp a skilleket is automatikusan felismeri a `.claude/skills/`-ből; a Claude Code plugin telepítése után a 8 agentmemory skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) natívan megjelenik a Warp slash-parancs palettáján. |
+| **Warp** | `~/.warp/.mcp.json` | Az `agentmemory connect warp` beírja a standard `mcpServers` blokkot. A Warp a skilleket is automatikusan felismeri a `.claude/skills/`-ből; a Claude Code plugin telepítése után a 9 agentmemory skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) natívan megjelenik a Warp slash-parancs palettáján. |
| **Cline (CLI)** | `~/.cline/mcp.json` | Az `agentmemory connect cline` beírja a standard `mcpServers` blokkot. VS Code extension felhasználóknak: illeszd be ugyanazt a blokkot a Cline Settings → MCP Servers → Edit JSON útján. |
| **Continue.dev** | `~/.continue/config.yaml` (előnyben részesített) vagy `config.json` (örökölt) | Az `agentmemory connect continue` létrehozza a hiányzó `config.yaml` fájlt, vagy módosítja a meglévő `config.json` fájlt. Meglévő YAML esetén kiírja az `mcpServers:` alá illeszthető blokkot, megőrizve a megjegyzéseket és horgonyokat. A Continue tömböt használ, nem objektumot. |
| **Zed** | `~/.config/zed/settings.json` | Az `agentmemory connect zed` a `context_servers` alá ír (a Zed saját kulcsa, NEM a `mcpServers`). A távoli MCP szerverek egy `{"url": "..."}`-vel is bekötődhetnek. |
@@ -925,7 +717,6 @@ Minden AI kódolóágens beépített memóriával érkezik: a Claude Code-nak `M
|---|---|---|
| Karbantartás | Kézzel gondozott instrukciók | Indexelt munkamenet-előzmények és emlékek |
| Keresés | Mindent betölt a kontextusba | BM25 + vektor + gráf (csak top-K) |
-| Token-költség | 22K+ 240 megfigyelésnél | ~1,900 token (92%-kal kevesebb) |
| Ágensek közötti | Ágensenkénti fájlok | MCP + REST (bármely ágens) |
| Koordináció | Nincs | Lease-ek, szignálok, akciók, rutinok |
| Megfigyelhetőség | Fájlok manuális olvasása | Valós idejű megjelenítő a :3113-on |
@@ -938,7 +729,7 @@ Minden AI kódolóágens beépített memóriával érkezik: a Claude Code-nak `M
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1023,7 +814,7 @@ Reciprocal Rank Fusionnal (RRF, k=60) fuzionálva, és sessionönkénti diverzif
Ha egy vektorindex fel van töltve, a `mem::search` (a `memory_recall` mögött) a hibrid BM25 + vektor rangsorolót használja. Embeddingek nélkül BM25-öt használ. A `smart-search` emellett strukturális gráfegyezéseket is fuzionálhat, ha létezik gráfadat, kulcs nélküli módban is. A tanulság-felidézés egy dedikált, memóriában tartott BM25-indexen fut, nem a teljes korpusz lekérdezésenkénti átnézésével. A felülírt memória-verziók minden felidézési útvonalból ki vannak zárva; a verziólánc megtartja az előzményüket.
-A vektorok túlélnek egy összeomlást vagy egy force-killt. A vektorindex legfeljebb minden `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`-ben (10 perc) kerül mentésre, csomagokban. Minden közben hozzáadott vagy eltávolított vektor azonnal felíródik egy kis, függőben lévő naplóba is az állapottárban, és a következő indítás lejátssza ezt az embedding-szolgáltató hívása nélkül. Minden sikeres mentés kiüríti a naplót. Azok a dokumentumok, amelyeknek a lejátszás után sincs vektoruk, a háttérben, `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) méretű batch-ekben kapnak újra embeddinget, amíg nem marad ilyen, és egy leállított backfill a következő indításkor folytatódik. A `/agentmemory/status` és a megjelenítő megmutatja a függőben lévő napló méretét és a backfill állapotát. A kulcs nélküli telepítések semmit nem írnak.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
A BM25 alapból tokenizálja a görögöt, a cirillt, a hébert, az arabot és az ékezetes latint. Kínai / japán / koreai memóriákhoz telepítsd az opcionális szegmentálókat (`npm install @node-rs/jieba tiny-segmenter`), hogy szóméretű tokenekre bontsa a CJK-futásokat; nélkülük az agentmemory finoman visszaesik a teljes futás tokenizálására, és egyszeri jelzést ír a stderr-re.
@@ -1039,9 +830,9 @@ A normál npm-telepítés tartalmazza az opcionális `@huggingface/transformers`
| Szolgáltató | Modell | Költség | Megjegyzések |
|---|---|---|---|
-| **Lokális (ajánlott opt-in)** | `all-MiniLM-L6-v2` | Ingyenes | Eszközön fut az első modell-letöltés után, +8pp felidézés a csak-BM25-höz képest |
+| **Lokális (ajánlott opt-in)** | `all-MiniLM-L6-v2` | Ingyenes | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Ingyenes szint | 100+ nyelv, 768/1536/3072 dimenzió (MRL), 2048 tokenes bemenet. Felváltja a `text-embedding-004`-et ([elavult, leállítva 2026. jan. 14-én](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Legjobb minőség |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Fizetős | Kódra optimalizálva |
| Cohere | `embed-english-v3.0` | Ingyenes próba | Általános célú |
| OpenRouter | Bármely modell | Változó | Több modellt lefedő proxy |
diff --git a/READMEs/README.id-ID.md b/READMEs/README.id-ID.md
index 1de7ed80..06a24851 100644
--- a/READMEs/README.id-ID.md
+++ b/READMEs/README.id-ID.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% token lebih sedikit" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 tool MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hook otomatis" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 DB eksternal" height="38" /></picture>
@@ -331,223 +329,17 @@ Catatan rilis terbaru: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmark" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Akurasi Retrieval
-
-**coding-agent-life-v1** (korpus in-house, bisa direproduksi di sandbox)
-
-| Adapter | P@5 | R@5 | Top-5 hit rate | Latensi p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Top-5 hit rate 100% pada **batas matematis P@5** untuk korpus ini (0.240, lihat scorecard). Hybrid me-retrieve setiap sesi gold; grep melewatkan 1 dari 2 gold pada query temporal multi-sesi. Peningkatannya ada pada **recall + temporal**, bukan presisi agregat. Benchmark ini kecil dan gold-nya jarang; LongMemEval-S yang lebih besar di bawah memberi pembeda yang lebih baik. Rincian lengkap per tipe + catatan koreksi: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 pertanyaan)
+Hasil penelusuran terkini, definisi metrik, dan langkah reproduksi:
-| Sistem | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Fallback BM25-only | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Penghematan Token
-
-| Pendekatan | Token/thn | Biaya/thn |
-|---|---|---|
-| Tempel konteks penuh | 19.5M+ | Tidak mungkin (melebihi window) |
-| Diringkas LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embedding lokal | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Model embedding: `all-MiniLM-L6-v2` (lokal, gratis, tanpa API key). Laporan lengkap: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Perbandingan kompetitor: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) mencakup agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproduksi secara lokal:** [`eval/README.md`](../eval/README.md), harness yang adapternya bisa dipasang-cabut untuk LongMemEval `_s` (500-Q publik) + `coding-agent-life-v1` (korpus in-house 15 sesi). Adapter grep / vector / agentmemory dinilai berdampingan, output NDJSON, scorecard yang dipublikasikan mendarat di [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Berpasangan dengan [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), dan [Graphify](https://github.com/safishamsi/graphify).** Code-graph indexing, pipeline build multi-agen, dan knowledge graph yang lebih luas lintas dokumen / PDF / gambar / video. agentmemory mengingat pekerjaannya; ketiga proyek itu menghidupkan sisa layer konteksnya. Resep + tabel perutean pertanyaan: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Kompetitor" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Bawaan (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tipe</strong></td>
-<td>Memory engine + server MCP</td>
-<td>API memory layer</td>
-<td>Runtime agen lengkap</td>
-<td>AI pribadi</td>
-<td>API memori + app</td>
-<td>Hub memori tim (proxy LLM)</td>
-<td>Vector memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Sistem memori</td>
-<td>File statis</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Self-reported</td>
-<td>PersonaMem 76% (self-reported)</td>
-<td>~96.6% (self-reported)</td>
-<td>94.4% (self-reported)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hook (tanpa effort manual)</td>
-<td>Panggilan <code>add()</code> manual</td>
-<td>Self-edit oleh agen</td>
-<td>Manual</td>
-<td>Ekstraksi di sisi API</td>
-<td>Intersepsi proxy (swap base-URL)</td>
-<td>Manual</td>
-<td>Ekstraksi API</td>
-<td>Manual</td>
-<td>Edit manual</td>
-</tr>
-<tr>
-<td><strong>Search</strong></td>
-<td>BM25 + Vector + Graph (fusi RRF)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semantik</td>
-<td>Vector + RAG</td>
-<td>4 tipe aset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vector-only</td>
-<td>Vector + semantik</td>
-<td>Decay-weighted</td>
-<td>Memuat semuanya ke konteks</td>
-</tr>
-<tr>
-<td><strong>Multi-agen</strong></td>
-<td>MCP + REST + lease + signal</td>
-<td>API (tanpa koordinasi)</td>
-<td>Hanya dalam runtime Letta</td>
-<td>Tidak</td>
-<td>Tidak</td>
-<td>Role tim + aset bersama</td>
-<td>Tidak</td>
-<td>Hanya terbatas (scoped)</td>
-<td>Multi-agen bersama</td>
-<td>File per agen</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>Tidak ada (klien MCP apa pun)</td>
-<td>Tidak ada</td>
-<td>Tinggi (harus pakai Letta)</td>
-<td>Standalone</td>
-<td>Tidak ada</td>
-<td>Proxy mengarahkan setiap panggilan model</td>
-<td>Tidak ada</td>
-<td>Oracle Database</td>
-<td>Tidak ada</td>
-<td>Format per agen</td>
-</tr>
-<tr>
-<td><strong>Dependensi eksternal</strong></td>
-<td>Tidak memerlukan layanan database terpisah</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>Multipel</td>
-<td>Cloud terkelola</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Tidak ada</td>
-<td>Tidak ada</td>
-</tr>
-<tr>
-<td><strong>Siklus hidup memori</strong></td>
-<td>Konsolidasi 4-tier + decay + auto-forget</td>
-<td>Ekstraksi pasif</td>
-<td>Dikelola agen</td>
-<td>Manual</td>
-<td>Auto-forget</td>
-<td>Review manual; auto-routing sedang dikerjakan</td>
-<td>Tidak ada</td>
-<td>Tidak disebutkan</td>
-<td>Decay + konsolidasi</td>
-<td>Pruning manual</td>
-</tr>
-<tr>
-<td><strong>Efisiensi token</strong></td>
-<td>~1.900 token/sesi ($10/thn)</td>
-<td>Bervariasi menurut integrasi</td>
-<td>Core memory di konteks</td>
-<td>Bervariasi</td>
-<td>Harga cloud</td>
-<td>Tidak disebutkan</td>
-<td>Tanpa budget token</td>
-<td>Didukung LLM (bervariasi)</td>
-<td>Bervariasi</td>
-<td>22K+ token pada 240 observasi</td>
-</tr>
-<tr>
-<td><strong>Viewer real-time</strong></td>
-<td>Ya (port 3113)</td>
-<td>Dashboard cloud</td>
-<td>Dashboard cloud</td>
-<td>Web UI</td>
-<td>Dashboard cloud</td>
-<td>Web UI Hub</td>
-<td>Tidak</td>
-<td>Tidak</td>
-<td>Tidak</td>
-<td>Tidak</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Ya (default)</td>
-<td>Opsional</td>
-<td>Opsional</td>
-<td>Ya</td>
-<td>Tidak (cloud-only)</td>
-<td>Ya (Docker)</td>
-<td>Ya</td>
-<td>Ya (Oracle DB)</td>
-<td>Ya</td>
-<td>Ya</td>
-</tr>
-</table>
-
-<sub>Catatan benchmark: hanya R@5 agentmemory yang merupakan hasil pengukuran kami sendiri (LongMemEval-S, bisa direproduksi dari <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Angka mem0 dan Letta adalah angka LoCoMo yang mereka publikasikan sendiri (dataset yang berbeda); angka MemPalace, supermemory, TencentDB (PersonaMem), dan oracleagentmemory adalah klaim self-reported vendor yang belum kami reproduksi secara independen (run oracleagentmemory memakai GPT-5.5 terhadap Oracle AI Database). Ditampilkan berdampingan hanya sebagai perkiraan kasar, bukan head-to-head pada data yang identik. Jumlah star bersifat perkiraan dan bisa berubah seiring waktu.</sub>
-
-**Pemain baru** yang layak diketahui, dibahas lebih dalam di [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistem | ⭐ | Sudut pandang |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Knowledge graph temporal; hasil query temporal terpublikasi terkuat (LongMemEval 63.8%), tapi graph dibangun secara asinkron sehingga fakta baru bisa tertinggal |
-| Cognee | 30K | Ingestion dokumen-ke-knowledge-graph, Python-only, dibangun untuk ekstraksi entitas terstruktur bukan capture sesi |
+Perbandingan produk dan opsi penerapan terkini, beserta sumbernya:
-AgentMemory memadukan capture sesi, riwayat proyek yang bisa dicari, dan viewer lokal, dengan mode tanpa kunci sejak awal.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Panduan lengkap: [`integrations/hermes/`](../integrations/hermes/)
Jalankan server memori: `npx -y @agentmemory/agentmemory@latest`
-#### Native skill lewat `npx skills add` (50+ agen)
+#### Native skill lewat `npx skills add`
agentmemory menyediakan 17 skill dalam format `<dir>/SKILL.md` ala Claude Code: 9 skill aksi yang bisa dipanggil (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) dan 8 skill referensi yang dimuat agen sesuai kebutuhan (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Skill referensi membawa tabel data yang dihasilkan dari source, sehingga tidak pernah drift. CLI [`skills`](https://npmjs.com/package/skills) oleh vercel-labs memasangnya secara otomatis ke direktori skill native agen yang memanggilnya, di 50+ agen (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, dan lainnya):
@@ -837,7 +629,7 @@ Entri agentmemory adalah **blok server MCP yang sama** di setiap host yang memak
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` memasang MCP dan hook capture di direktori kustomisasi bersama. Lihat [pengaturan dan batasan Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` memakai konfigurasi MCP dan hook yang sama dengan versi IDE saat ini. Instalasi yang sudah ada sebaiknya di-refresh dengan `--force`; lihat [catatan upgrade](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` menulis konfigurasi tingkat pengguna. Override workspace masuk di `.kiro/settings/mcp.json` di sebelah kode Anda. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` menulis blok `mcpServers` standar. Warp juga auto-discover skill dari `.claude/skills/`; setelah plugin Claude Code terpasang, 8 skill agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) muncul secara native di palet slash-command Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` menulis blok `mcpServers` standar. Warp juga auto-discover skill dari `.claude/skills/`; setelah plugin Claude Code terpasang, 9 skill agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) muncul secara native di palet slash-command Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` menulis blok `mcpServers` standar. Pengguna ekstensi VS Code: tempel blok yang sama lewat Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (disarankan) atau `config.json` (legacy) | `agentmemory connect continue` membuat `config.yaml` jika belum ada konfigurasi atau memperbarui `config.json` yang ada. YAML yang sudah ada dipertahankan agar komentar dan anchor tetap utuh; perintah mencetak blok untuk ditempel di bawah `mcpServers:`. Continue memakai bentuk array untuk `mcpServers`. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` menulis di bawah `context_servers` (key milik Zed, BUKAN `mcpServers`). Server MCP remote bisa disambungkan lewat `{"url": "..."}` sebagai gantinya. |
@@ -934,7 +726,6 @@ Setiap agen coding AI dilengkapi memori bawaan: Claude Code punya `MEMORY.md`, C
|---|---|---|
| Pemeliharaan | Instruksi yang dikelola manual | Riwayat sesi dan memori terindeks |
| Search | Memuat semuanya ke konteks | BM25 + vector + graph (hanya top-K) |
-| Biaya token | 22K+ pada 240 observasi | ~1.900 token (92% lebih sedikit) |
| Lintas agen | File per agen | MCP + REST (agen apa pun) |
| Koordinasi | Tidak ada | Lease, signal, action, routine |
| Observability | Baca file secara manual | Viewer real-time di :3113 |
@@ -947,7 +738,7 @@ Setiap agen coding AI dilengkapi memori bawaan: Claude Code punya `MEMORY.md`, C
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1032,7 +823,7 @@ Difusikan dengan Reciprocal Rank Fusion (RRF, k=60) dan didiversifikasi per sesi
Saat index vector terisi, `mem::search` (di belakang `memory_recall`) memakai ranker hybrid BM25 + vector. Tanpa embedding ia memakai BM25. `smart-search` juga bisa menggabungkan kecocokan graph struktural saat data graph ada, termasuk di mode keyless. Recall lesson berjalan di index BM25 in-memory khusus bukan memindai seluruh korpus per query. Versi memori yang disupersede dikecualikan dari setiap jalur recall; rantai versi menyimpan histori mereka.
-Vector tetap bertahan dari crash atau force-kill. Index vector disimpan dalam bucket paling lambat setiap `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 menit). Setiap vector yang ditambahkan atau dihapus di antaranya juga langsung ditulis ke log pending kecil di state store, dan start berikutnya me-replay-nya tanpa memanggil penyedia embedding. Setiap save yang berhasil mengosongkan log itu. Dokumen yang masih belum punya vector setelah replay di-embed ulang di background dalam batch `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) sampai tidak ada yang tersisa, dan backfill yang terhenti dilanjutkan pada start berikutnya. `/agentmemory/status` dan viewer menunjukkan ukuran log pending dan state backfill. Instalasi keyless tidak menulis apa pun.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 men-tokenisasi Yunani, Sirilik, Ibrani, Arab, dan Latin beraksen secara bawaan. Untuk memori Tionghoa / Jepang / Korea, pasang segmenter opsional (`npm install @node-rs/jieba tiny-segmenter`) untuk membelah runtun CJK menjadi token level-kata; tanpa itu, agentmemory soft-fallback ke tokenisasi seluruh runtun dan mencetak hint satu-kali di stderr.
@@ -1048,9 +839,9 @@ Instalasi npm normal menyertakan runtime opsional `@huggingface/transformers`. P
| Penyedia | Model | Biaya | Catatan |
|---|---|---|---|
-| **Lokal (opt-in yang disarankan)** | `all-MiniLM-L6-v2` | Gratis | Di perangkat sendiri setelah unduhan model pertama, +8pp recall dibanding BM25-only |
+| **Lokal (opt-in yang disarankan)** | `all-MiniLM-L6-v2` | Gratis | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free tier | 100+ bahasa, 768/1536/3072 dimensi (MRL), input 2048-token. Menggantikan `text-embedding-004` ([deprecated, shutdown 14 Jan 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Kualitas tertinggi |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Berbayar | Dioptimalkan untuk code |
| Cohere | `embed-english-v3.0` | Free trial | Tujuan umum |
| OpenRouter | Model apa pun | Bervariasi | Proxy multi-model |
diff --git a/READMEs/README.it-IT.md b/READMEs/README.it-IT.md
index 3e3fa649..871d4093 100644
--- a/READMEs/README.it-IT.md
+++ b/READMEs/README.it-IT.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -331,223 +329,17 @@ Note sull'ultima release: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmark" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Precisione del retrieval
-
-**coding-agent-life-v1** (corpus interno, riproducibile in sandbox)
-
-| Adapter | P@5 | R@5 | Tasso di successo top-5 | latenza p50 |
-|---|---|---|---|---|
-| **agentmemory ibrido** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep di base | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Tasso di successo top-5 del 100% al **limite matematico di P@5** per questo corpus (0.240, vedi scorecard). L'ibrido recupera ogni sessione gold; grep ne manca 1 su 2 nella query temporale multi-sessione. Il guadagno è su **recall + temporale**, non sulla precisione aggregata. Questo benchmark è piccolo e povero di dati gold; il più ampio LongMemEval-S sotto differenzia meglio. Analisi completa per tipo + nota di correzione: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 domande)
+Risultati attuali del recupero, definizioni delle metriche e procedura di riproduzione:
-| Sistema | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Fallback solo BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Risparmio di token
-
-| Approccio | Token/anno | Costo/anno |
-|---|---|---|
-| Incolla il contesto completo | 19.5M+ | Impossibile (eccede la finestra) |
-| Riassunto da LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embedding locali | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Modello di embedding: `all-MiniLM-L6-v2` (locale, gratuito, senza chiave API). Report completi: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Confronto con i concorrenti: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) che copre agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Riproducibile in locale:** [`eval/README.md`](../eval/README.md), un harness con adapter collegabili per LongMemEval `_s` (500 domande pubbliche) + `coding-agent-life-v1` (corpus interno di 15 sessioni). Gli adapter grep / vettoriale / agentmemory ottengono un punteggio fianco a fianco, output NDJSON, le scorecard pubblicate finiscono in [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Si combina con [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) e [Graphify](https://github.com/safishamsi/graphify).** Indicizzazione code-graph, pipeline di build multi-agente e grafi di conoscenza più ampi su documenti / PDF / immagini / video. agentmemory ricorda il lavoro; questi tre progetti illuminano il resto dello strato di contesto. Ricette + tabella di instradamento delle domande: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Concorrenti" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Integrata (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tipo</strong></td>
-<td>Motore di memoria + server MCP</td>
-<td>API livello memoria</td>
-<td>Runtime agente completo</td>
-<td>IA personale</td>
-<td>API memoria + app</td>
-<td>Hub memoria di team (proxy LLM)</td>
-<td>Memoria vettoriale (OSS)</td>
-<td>Motore di memoria (Oracle DB)</td>
-<td>Sistema di memoria</td>
-<td>File statico</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/D</td>
-<td>Auto-dichiarato</td>
-<td>PersonaMem 76% (auto-dichiarato)</td>
-<td>~96.6% (auto-dichiarato)</td>
-<td>94.4% (auto-dichiarato)</td>
-<td>N/D</td>
-<td>N/D (grep)</td>
-</tr>
-<tr>
-<td><strong>Acquisizione automatica</strong></td>
-<td>12 hook (zero sforzo manuale)</td>
-<td>Chiamate manuali <code>add()</code></td>
-<td>Auto-modifiche dell'agente</td>
-<td>Manuale</td>
-<td>Estrazione lato API</td>
-<td>Intercettazione proxy (sostituzione base-URL)</td>
-<td>Manuale</td>
-<td>Estrazione API</td>
-<td>Manuale</td>
-<td>Modifica manuale</td>
-</tr>
-<tr>
-<td><strong>Ricerca</strong></td>
-<td>BM25 + vettoriale + grafo (fusione RRF)</td>
-<td>Vettoriale + grafo</td>
-<td>Vettoriale (archivio)</td>
-<td>Semantica</td>
-<td>Vettoriale + RAG</td>
-<td>4 tipi di asset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Solo vettoriale</td>
-<td>Vettoriale + semantica</td>
-<td>Pesata per decadimento</td>
-<td>Carica tutto nel contesto</td>
-</tr>
-<tr>
-<td><strong>Multi-agente</strong></td>
-<td>MCP + REST + lease + segnali</td>
-<td>API (nessun coordinamento)</td>
-<td>Solo all'interno del runtime Letta</td>
-<td>No</td>
-<td>No</td>
-<td>Ruoli di team + asset condivisi</td>
-<td>No</td>
-<td>Solo con ambito limitato</td>
-<td>Condiviso multi-agente</td>
-<td>File per agente</td>
-</tr>
-<tr>
-<td><strong>Vincolo al framework</strong></td>
-<td>Nessuno (qualsiasi client MCP)</td>
-<td>Nessuno</td>
-<td>Alto (deve usare Letta)</td>
-<td>Autonomo</td>
-<td>Nessuno</td>
-<td>Il proxy intercetta ogni chiamata al modello</td>
-<td>Nessuno</td>
-<td>Oracle Database</td>
-<td>Nessuno</td>
-<td>Formato per agente</td>
-</tr>
-<tr>
-<td><strong>Dipendenze esterne</strong></td>
-<td>Nessun servizio database separato richiesto</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + DB vettoriale</td>
-<td>Multiple</td>
-<td>Cloud gestito</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Nessuna</td>
-<td>Nessuna</td>
-</tr>
-<tr>
-<td><strong>Ciclo di vita della memoria</strong></td>
-<td>Consolidamento a 4 livelli + decadimento + auto-oblio</td>
-<td>Estrazione passiva</td>
-<td>Gestito dall'agente</td>
-<td>Manuale</td>
-<td>Auto-oblio</td>
-<td>Revisione manuale; instradamento automatico in corso</td>
-<td>Nessuno</td>
-<td>Non dichiarato</td>
-<td>Decadimento + consolidamento</td>
-<td>Pulizia manuale</td>
-</tr>
-<tr>
-<td><strong>Efficienza dei token</strong></td>
-<td>~1,900 token/sessione ($10/anno)</td>
-<td>Varia per integrazione</td>
-<td>Memoria core nel contesto</td>
-<td>Varia</td>
-<td>Prezzo cloud</td>
-<td>Non dichiarato</td>
-<td>Nessun budget di token</td>
-<td>Basato su LLM (varia)</td>
-<td>Varia</td>
-<td>22K+ token a 240 osservazioni</td>
-</tr>
-<tr>
-<td><strong>Viewer in tempo reale</strong></td>
-<td>Sì (porta 3113)</td>
-<td>Dashboard cloud</td>
-<td>Dashboard cloud</td>
-<td>UI web</td>
-<td>Dashboard cloud</td>
-<td>UI web Hub</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-<td>No</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Sì (predefinito)</td>
-<td>Opzionale</td>
-<td>Opzionale</td>
-<td>Sì</td>
-<td>No (solo cloud)</td>
-<td>Sì (Docker)</td>
-<td>Sì</td>
-<td>Sì (Oracle DB)</td>
-<td>Sì</td>
-<td>Sì</td>
-</tr>
-</table>
-
-<sub>Nota sul benchmark: solo l'R@5 di agentmemory è un risultato misurato da noi (LongMemEval-S, riproducibile da <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). I valori di mem0 e Letta sono i loro numeri pubblicati su LoCoMo (un dataset diverso); i valori di MemPalace, supermemory, TencentDB (PersonaMem) e oracleagentmemory sono dichiarazioni auto-riportate dai rispettivi fornitori che non abbiamo riprodotto in modo indipendente (l'esecuzione di oracleagentmemory ha usato GPT-5.5 contro un Oracle AI Database). Mostrati fianco a fianco solo per un confronto approssimativo, non un testa a testa su dati identici. I conteggi delle stelle sono approssimativi e cambiano nel tempo.</sub>
-
-**Nuovi arrivati** da conoscere, confrontati in profondità in [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistema | ⭐ | Approccio |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Grafo di conoscenza temporale; i risultati pubblicati più solidi sulle query temporali (LongMemEval 63.8%), ma il grafo si costruisce in modo asincrono quindi i fatti recenti possono essere in ritardo |
-| Cognee | 30K | Ingestione da documento a grafo di conoscenza, solo Python, costruito per l'estrazione strutturata di entità più che per l'acquisizione di sessioni |
+Confronto aggiornato dei prodotti e delle opzioni di distribuzione, con fonti:
-AgentMemory combina acquisizione dagli agenti di programmazione, cronologia del progetto ricercabile e visualizzatore locale. La modalità senza chiave è disponibile dal primo avvio.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Guida completa: [`integrations/hermes/`](../integrations/hermes/)
Avvia il server di memoria: `npx -y @agentmemory/agentmemory@latest`
-#### Skill native tramite `npx skills add` (50+ agenti)
+#### Skill native tramite `npx skills add`
agentmemory distribuisce 17 skill nel formato `<dir>/SKILL.md` in stile Claude Code: 9 skill d'azione invocabili (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) e 8 skill di riferimento che l'agente carica su richiesta (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Le skill di riferimento portano tabelle di dati generate dalla fonte, quindi non diventano mai obsolete. La CLI [`skills`](https://npmjs.com/package/skills) di vercel-labs le installa automaticamente nella directory delle skill native dell'agente chiamante su 50+ agenti (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, e altri):
@@ -836,7 +628,7 @@ La voce agentmemory è lo **stesso blocco server MCP** su ogni host che usa la f
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installa l'MCP e gli hook di acquisizione nella directory di personalizzazione condivisa. Consulta [configurazione e limiti di Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` usa la stessa configurazione MCP e hook delle versioni IDE attuali. Le installazioni esistenti dovrebbero aggiornare con `--force`; consulta le [note di aggiornamento](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` scrive la configurazione a livello utente. Gli override per workspace vanno in `.kiro/settings/mcp.json` accanto al tuo codice. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` scrive il blocco standard `mcpServers`. Warp rileva automaticamente anche le skill da `.claude/skills/`; una volta installato il plugin Claude Code, le 8 skill di agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) appaiono nativamente nella palette slash-command di Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` scrive il blocco standard `mcpServers`. Warp rileva automaticamente anche le skill da `.claude/skills/`; una volta installato il plugin Claude Code, le 9 skill di agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) appaiono nativamente nella palette slash-command di Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` scrive il blocco standard `mcpServers`. Utenti dell'estensione VS Code: incolla lo stesso blocco tramite Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferito) o `config.json` (legacy) | `agentmemory connect continue` crea `config.yaml` se manca o modifica il `config.json` esistente. Se il file YAML esiste, mostra il blocco da aggiungere a `mcpServers:` per conservare commenti e ancore. Continue usa un array, non un oggetto. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` scrive sotto `context_servers` (la chiave di Zed, NON `mcpServers`). I server MCP remoti possono essere collegati tramite `{"url": "..."}` invece. |
@@ -933,7 +725,6 @@ Ogni agente di coding AI ha una memoria integrata: Claude Code ha `MEMORY.md`, C
|---|---|---|
| Manutenzione | Istruzioni curate manualmente | Cronologia delle sessioni e ricordi indicizzati |
| Ricerca | Carica tutto nel contesto | BM25 + vettoriale + grafo (solo top-K) |
-| Costo in token | 22K+ a 240 osservazioni | ~1,900 token (92% in meno) |
| Cross-agente | File per agente | MCP + REST (qualsiasi agente) |
| Coordinamento | Nessuno | Lease, segnali, azioni, routine |
| Osservabilità | Lettura manuale dei file | Viewer in tempo reale su :3113 |
@@ -946,7 +737,7 @@ Ogni agente di coding AI ha una memoria integrata: Claude Code ha `MEMORY.md`, C
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Fuso con Reciprocal Rank Fusion (RRF, k=60) e diversificato per sessione (massim
Quando un indice vettoriale è popolato, `mem::search` (dietro `memory_recall`) usa il ranker ibrido BM25 + vettoriale. Senza embedding usa BM25. `smart-search` può inoltre fondere corrispondenze strutturali del grafo quando i dati del grafo esistono, anche in modalità keyless. Il recall delle lezioni funziona su un indice BM25 dedicato in memoria invece di scansionare l'intero corpus per ogni query. Le versioni superate della memoria sono escluse da ogni percorso di recall; la catena delle versioni conserva la loro storia.
-I vettori sopravvivono a un crash o a un force-kill. L'indice vettoriale viene salvato a blocchi almeno ogni `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minuti). Ogni vettore aggiunto o rimosso nel frattempo viene anche scritto immediatamente in un piccolo log in sospeso nello state store, e il prossimo avvio lo riapplica senza chiamare il provider di embedding. Ogni salvataggio riuscito svuota il log. I documenti che restano senza vettore dopo la riapplicazione vengono ri-embeddati in background a lotti di `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) finché non ne resta nessuno, e un backfill interrotto riprende al prossimo avvio. `/agentmemory/status` e il viewer mostrano la dimensione del log in sospeso e lo stato del backfill. Le installazioni keyless non scrivono nulla.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenizza nativamente greco, cirillico, ebraico, arabo e latino accentato. Per le memorie in cinese / giapponese / coreano, installa i segmentatori opzionali (`npm install @node-rs/jieba tiny-segmenter`) per dividere le sequenze CJK in token a livello di parola; senza di essi, agentmemory ricade in modo graduale sulla tokenizzazione dell'intera sequenza e stampa un suggerimento una tantum su stderr.
@@ -1047,9 +838,9 @@ L'installazione npm normale include il runtime opzionale `@huggingface/transform
| Provider | Modello | Costo | Note |
|---|---|---|---|
-| **Locale (consigliato come opt-in)** | `all-MiniLM-L6-v2` | Gratuito | On-device dopo il primo download del modello, +8pp di recall rispetto a solo BM25 |
+| **Locale (consigliato come opt-in)** | `all-MiniLM-L6-v2` | Gratuito | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Livello gratuito | 100+ lingue, 768/1536/3072 dimensioni (MRL), input di 2048 token. Sostituisce `text-embedding-004` ([deprecato, spegnimento il 14 gennaio 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Qualità più alta |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | A pagamento | Ottimizzato per il codice |
| Cohere | `embed-english-v3.0` | Prova gratuita | Uso generale |
| OpenRouter | Qualsiasi modello | Variabile | Proxy multi-modello |
diff --git a/READMEs/README.ja-JP.md b/READMEs/README.ja-JP.md
index 12b8352c..050f4dbb 100644
--- a/READMEs/README.ja-JP.md
+++ b/READMEs/README.ja-JP.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="検索 R@5 95.2%" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="トークン 92% 削減" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP ツール" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 自動 hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="外部 DB 0 個" height="38" /></picture>
@@ -331,223 +329,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### 検索精度
-
-**coding-agent-life-v1**(社内コーパス、サンドボックスで再現可能)
-
-| アダプタ | P@5 | R@5 | Top-5 ヒット率 | p50 レイテンシ |
-|---|---|---|---|---|
-| **agentmemory ハイブリッド** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep ベースライン | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-このコーパスの **P@5 の数学的上限**(0.240、スコアカード参照)で Top-5 ヒット率 100% を達成。ハイブリッドはすべてのゴールドセッションを取得しますが、grep は複数セッションにわたる時系列クエリでゴールドの 2 件中 1 件を取りこぼします。向上しているのは**リコール + 時系列性**であり、集計精度ではありません。このベンチマークは小規模でゴールドが少なく、下の、より大きな LongMemEval-S のほうがより明確に差を示します。種類別の詳細な内訳と補正の注記: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)。
-
-**LongMemEval-S**(ICLR 2025、500 問)
+最新の検索評価結果、指標の定義、再現手順:
-| システム | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25 のみフォールバック | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### トークン削減
-
-| アプローチ | トークン/年 | コスト/年 |
-|---|---|---|
-| フルコンテキストを貼り付け | 19.5M+ | 不可能(ウィンドウを超過) |
-| LLM による要約 | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + ローカル埋め込み | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> 埋め込みモデル: `all-MiniLM-L6-v2`(ローカル、無料、API キー不要)。完全なレポート: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md)、[`benchmark/QUALITY.md`](../benchmark/QUALITY.md)、[`benchmark/SCALE.md`](../benchmark/SCALE.md)。競合比較: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) では agentmemory と mem0、Letta、Khoj、supermemory、TencentDB Agent Memory、MemPalace、Zep/Graphiti、Cognee、Hippo を比較しています。
-
-**ローカルで再現する:** [`eval/README.md`](../eval/README.md) は LongMemEval `_s`(公開 500 問)と `coding-agent-life-v1`(社内 15 セッションのコーパス)向けのアダプタ差し替え可能なハーネスです。grep / ベクトル / agentmemory の各アダプタを横並びでスコアリングし、NDJSON を出力し、公開済みのスコアカードは [`docs/benchmarks/`](../docs/benchmarks/) に置かれます。
-
-**[codegraph](https://github.com/colbymchenry/codegraph)、[Understand Anything](https://github.com/Lum1104/Understand-Anything)、[Graphify](https://github.com/safishamsi/graphify) と組み合わせる。** コードグラフのインデックス化、マルチエージェントのビルドパイプライン、ドキュメント / PDF / 画像 / 動画をまたぐより広いナレッジグラフ。agentmemory は作業内容を記憶し、この 3 つのプロジェクトはコンテキスト層の残りの部分を照らします。レシピと質問のルーティング表: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md)。
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>組み込み (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>種別</strong></td>
-<td>メモリエンジン + MCP サーバー</td>
-<td>メモリレイヤー API</td>
-<td>フルエージェントランタイム</td>
-<td>パーソナル AI</td>
-<td>メモリ API + アプリ</td>
-<td>チームメモリハブ(LLM プロキシ)</td>
-<td>ベクトルメモリ(OSS)</td>
-<td>メモリエンジン(Oracle DB)</td>
-<td>メモリシステム</td>
-<td>静的ファイル</td>
-</tr>
-<tr>
-<td><strong>検索 R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>自己申告</td>
-<td>PersonaMem 76%(自己申告)</td>
-<td>~96.6%(自己申告)</td>
-<td>94.4%(自己申告)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>自動キャプチャ</strong></td>
-<td>12 hooks(手動作業ゼロ)</td>
-<td>手動の <code>add()</code> 呼び出し</td>
-<td>エージェントが自分で編集</td>
-<td>手動</td>
-<td>API 側での抽出</td>
-<td>プロキシ横取り(base-URL 差し替え)</td>
-<td>手動</td>
-<td>API 抽出</td>
-<td>手動</td>
-<td>手動編集</td>
-</tr>
-<tr>
-<td><strong>検索</strong></td>
-<td>BM25 + ベクトル + グラフ(RRF 融合)</td>
-<td>ベクトル + グラフ</td>
-<td>ベクトル(アーカイブ)</td>
-<td>セマンティック</td>
-<td>ベクトル + RAG</td>
-<td>4 種のアセット(Chat / Skill / Wiki / CodeGraph)</td>
-<td>ベクトルのみ</td>
-<td>ベクトル + セマンティック</td>
-<td>減衰重み付け</td>
-<td>すべてをコンテキストにロード</td>
-</tr>
-<tr>
-<td><strong>マルチエージェント</strong></td>
-<td>MCP + REST + リース + シグナル</td>
-<td>API(調整なし)</td>
-<td>Letta ランタイム内のみ</td>
-<td>なし</td>
-<td>なし</td>
-<td>チームロール + 共有アセット</td>
-<td>なし</td>
-<td>スコープのみ</td>
-<td>マルチエージェント共有</td>
-<td>エージェントごとのファイル</td>
-</tr>
-<tr>
-<td><strong>フレームワークロックイン</strong></td>
-<td>なし(任意の MCP クライアント)</td>
-<td>なし</td>
-<td>高(Letta 必須)</td>
-<td>スタンドアロン</td>
-<td>なし</td>
-<td>プロキシがすべてのモデル呼び出しを仲介</td>
-<td>なし</td>
-<td>Oracle Database</td>
-<td>なし</td>
-<td>エージェントごとのフォーマット</td>
-</tr>
-<tr>
-<td><strong>外部依存</strong></td>
-<td>別のデータベースサービスは不要</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + ベクトル DB</td>
-<td>複数</td>
-<td>マネージドクラウド</td>
-<td>Docker スタック(Core + Hub + Proxy)</td>
-<td>ベクトルストア</td>
-<td>Oracle AI Database</td>
-<td>なし</td>
-<td>なし</td>
-</tr>
-<tr>
-<td><strong>メモリライフサイクル</strong></td>
-<td>4 層統合 + 減衰 + 自動忘却</td>
-<td>受動的抽出</td>
-<td>エージェント管理</td>
-<td>手動</td>
-<td>自動忘却</td>
-<td>手動レビュー(自動ルーティングは開発中)</td>
-<td>なし</td>
-<td>記載なし</td>
-<td>減衰 + 統合</td>
-<td>手動プルーニング</td>
-</tr>
-<tr>
-<td><strong>トークン効率</strong></td>
-<td>~1,900 tokens/セッション ($10/年)</td>
-<td>統合方法による</td>
-<td>コアメモリがコンテキスト内</td>
-<td>場合による</td>
-<td>クラウド価格</td>
-<td>記載なし</td>
-<td>トークン予算なし</td>
-<td>LLM ベース(場合による)</td>
-<td>場合による</td>
-<td>240 観測で 22K+ tokens</td>
-</tr>
-<tr>
-<td><strong>リアルタイムビューワー</strong></td>
-<td>あり(ポート 3113)</td>
-<td>クラウドダッシュボード</td>
-<td>クラウドダッシュボード</td>
-<td>Web UI</td>
-<td>クラウドダッシュボード</td>
-<td>Hub の Web UI</td>
-<td>なし</td>
-<td>なし</td>
-<td>なし</td>
-<td>なし</td>
-</tr>
-<tr>
-<td><strong>セルフホスト</strong></td>
-<td>あり(デフォルト)</td>
-<td>オプション</td>
-<td>オプション</td>
-<td>あり</td>
-<td>なし(クラウドのみ)</td>
-<td>あり(Docker)</td>
-<td>あり</td>
-<td>あり(Oracle DB)</td>
-<td>あり</td>
-<td>あり</td>
-</tr>
-</table>
-
-<sub>ベンチマークに関する注記: 自社で計測した結果は agentmemory の R@5 のみです(LongMemEval-S、<a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> から再現可能)。mem0 と Letta の数値は各社が公表した LoCoMo の数値(別のデータセット)です。MemPalace、supermemory、TencentDB(PersonaMem)、oracleagentmemory の数値はベンダーの自己申告で、当方では独立に再現していません(oracleagentmemory の実行は Oracle AI Database に対して GPT-5.5 を使用しています)。並べて示しているのはあくまで目安であり、同一データでの直接対決ではありません。スター数は概数で、時間とともに変動します。</sub>
-
-**知っておく価値のある新顔**で、[`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) で詳しく比較しています:
-
-| システム | ⭐ | 切り口 |
-|--------|---|-------|
-| Zep / Graphiti | 30K | 時系列ナレッジグラフ。公表されている時系列クエリの結果としては最強(LongMemEval 63.8%)だが、グラフ構築が非同期のため新しい事実の反映が遅れることがある |
-| Cognee | 30K | ドキュメントからナレッジグラフへの取り込み。Python のみで、セッションキャプチャではなく構造化エンティティ抽出のために作られている |
+製品と導入方法の最新比較(出典付き):
-AgentMemory はセッションのキャプチャ、検索可能なプロジェクト履歴、ローカルビューアーを組み合わせ、初回からキーレスで使えます。
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
メモリサーバーを起動: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add` によるネイティブ skills(50+ エージェント)
+#### `npx skills add` によるネイティブ skills
agentmemory は Claude Code スタイルの `<dir>/SKILL.md` フォーマットで 17 個の skills を同梱しています: 呼び出し可能な 9 つのアクション skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)と、エージェントがオンデマンドで読み込む 8 個のリファレンス skills(`memory-discipline`、`agentmemory-mcp-tools`、`agentmemory-rest-api`、`agentmemory-config`、`agentmemory-agents`、`agentmemory-hooks`、`agentmemory-architecture`、`write-agentmemory-skill`)です。リファレンス skills はソースから生成されたデータ表を含むため、決してドリフトしません。vercel-labs による [`skills`](https://npmjs.com/package/skills) CLI が、呼び出し元エージェントのネイティブ skill ディレクトリへ 50 以上のエージェント(Claude Code、Cursor、Cline、Continue、Droid、Warp、Codex、Antigravity、Kiro、OpenCode、Goose、Roo、Trae、Windsurf など)にわたって自動インストールします:
@@ -837,7 +629,7 @@ agentmemory のエントリは、`mcpServers` の形式を使うすべてのホ
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` が共有のカスタマイズディレクトリに MCP とキャプチャ hooks をインストールします。[Antigravity のセットアップと制限](../docs/plugins/antigravity.md)を参照。 |
| **Antigravity CLI**(`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` は現在の IDE バージョンと同じ MCP と hook の設定を使います。既存のインストールは `--force` で更新してください。[アップグレードノート](../docs/plugins/antigravity.md)を参照。 |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` がユーザーレベルの設定を書き込みます。ワークスペース単位の上書きは、コードの隣にある `.kiro/settings/mcp.json` に書いてください。 |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` が標準の `mcpServers` ブロックを書き込みます。Warp は `.claude/skills/` からも skills を自動検出します。Claude Code プラグインをインストールすれば、8 つの agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`commit-context`、`commit-history`、`session-history`)が Warp のスラッシュコマンドパレットにネイティブで表示されます。 |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` が標準の `mcpServers` ブロックを書き込みます。Warp は `.claude/skills/` からも skills を自動検出します。Claude Code プラグインをインストールすれば、9 つの agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)が Warp のスラッシュコマンドパレットにネイティブで表示されます。 |
| **Cline(CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` が標準の `mcpServers` ブロックを書き込みます。VS Code 拡張を使う場合は、Cline Settings → MCP Servers → Edit JSON に同じブロックを貼り付けてください。 |
| **Continue.dev** | `~/.continue/config.yaml`(推奨)または `config.json`(レガシー) | `agentmemory connect continue` は設定がなければ `config.yaml` を作成し、既存の `config.json` は更新します。既存の YAML はコメントとアンカーを維持するため書き換えず、`mcpServers:` に貼るブロックを表示します。Continue の `mcpServers` は配列です。 |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` は `context_servers`(`mcpServers` ではなく Zed 独自のキー)の下に書き込みます。リモート MCP サーバーは、代わりに `{"url": "..."}` で配線できます。 |
@@ -933,7 +725,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| 保守 | 手動で管理する指示 | インデックス化されたセッション履歴とメモリ |
| 検索 | すべてをコンテキストにロード | BM25 + ベクトル + グラフ(top-K のみ) |
-| トークンコスト | 240 観測で 22K+ | ~1,900 トークン(92% 削減) |
| クロスエージェント | エージェントごとのファイル | MCP + REST(任意のエージェント) |
| 調整 | なし | リース、シグナル、アクション、ルーチン |
| 可観測性 | 手動でファイルを読む | ポート 3113 のリアルタイムビューワー |
@@ -946,7 +737,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Reciprocal Rank Fusion(RRF、k=60)で融合し、セッションごとに多様
ベクトルインデックスが構築されている場合、`mem::search`(`memory_recall` の背後)はハイブリッドな BM25 + ベクトルランカーを使います。埋め込みがない場合は BM25 を使います。`smart-search` は、キーレスモードでも、グラフデータが存在する場合は構造的なグラフマッチをさらに融合できます。レッスンのリコールは、クエリごとにコーパス全体をスキャンするのではなく、専用のインメモリ BM25 インデックス上で動作します。上書きされたメモリバージョンはすべてのリコール経路から除外されます。バージョンチェーンがその履歴を保持します。
-ベクトルはクラッシュや強制終了を生き延びます。ベクトルインデックスは、少なくとも `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`(10 分)ごとにバケット単位で保存されます。その間に追加・削除されたベクトルはすべて、ステートストア内の小さな保留ログにも即座に書き込まれ、次回起動時にそれが埋め込みプロバイダーを呼び出さずに再生されます。保存が成功するたびにそのログは空になります。再生後もベクトルを持たないドキュメントは、何も残らなくなるまで `AGENTMEMORY_VECTOR_BACKFILL_MAX`(500)件のバッチでバックグラウンドで再埋め込みされ、中断されたバックフィルは次回起動時に続行されます。`/agentmemory/status` とビューワーは、保留ログのサイズとバックフィルの状態を表示します。キーレスインストールは何も書き込みません。
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 は、ギリシャ語、キリル文字、ヘブライ語、アラビア語、アクセント付きラテン文字をそのままトークン化します。中国語 / 日本語 / 韓国語のメモリについては、オプションのセグメンタ(`npm install @node-rs/jieba tiny-segmenter`)をインストールすると、CJK の文字列を単語レベルのトークンに分割できます。インストールしない場合、agentmemory は文字列全体をそのままトークン化するようソフトにフォールバックし、stderr に 1 回だけヒントを表示します。
@@ -1047,9 +838,9 @@ EMBEDDING_PROVIDER=local
| プロバイダー | モデル | コスト | 備考 |
|---|---|---|---|
-| **ローカル(推奨オプトイン)** | `all-MiniLM-L6-v2` | 無料 | 最初のモデルダウンロード後はオンデバイス。BM25 のみに比べて R@5 +8pp |
+| **ローカル(推奨オプトイン)** | `all-MiniLM-L6-v2` | 無料 | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | 無料枠 | 100+ 言語、768/1536/3072 次元(MRL)、2048 トークン入力。`text-embedding-004`([非推奨、2026 年 1 月 14 日に終了](https://ai.google.dev/gemini-api/docs/deprecations))の後継 |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | 最高品質 |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | 有料 | コードに最適化 |
| Cohere | `embed-english-v3.0` | 無料トライアル | 汎用 |
| OpenRouter | 任意のモデル | 様々 | マルチモデルプロキシ |
diff --git a/READMEs/README.ko-KR.md b/READMEs/README.ko-KR.md
index 37e27d89..2ba26f74 100644
--- a/READMEs/README.ko-KR.md
+++ b/READMEs/README.ko-KR.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -329,223 +327,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="벤치마크" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### 검색 정확도
-
-**coding-agent-life-v1** (자체 코퍼스, 샌드박스 재현 가능)
-
-| 어댑터 | P@5 | R@5 | Top-5 적중률 | p50 지연 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-이 코퍼스의 **P@5 수학적 상한**(0.240, 스코어카드 참고)에서 Top-5 적중률 100%. 하이브리드는 모든 gold 세션을 검색하지만, grep은 멀티 세션 시간 쿼리에서 gold 2개 중 1개를 놓칩니다. 이득은 종합 정밀도가 아니라 **리콜 + 시간성**입니다. 이 벤치마크는 작고 gold가 희소하며, 아래의 더 큰 LongMemEval-S가 더 잘 변별합니다. 유형별 전체 분석 + 정정 노트: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500개 질문)
+최신 검색 평가 결과, 지표 정의 및 재현 절차:
-| 시스템 | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### 토큰 절감
-
-| 방식 | 연간 토큰 | 연간 비용 |
-|---|---|---|
-| 전체 컨텍스트를 매번 붙여넣기 | 19.5M+ | 불가능(컨텍스트 윈도우 초과) |
-| LLM 요약 | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + 로컬 임베딩 | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> 임베딩 모델: `all-MiniLM-L6-v2` (로컬, 무료, API 키 불필요). 전체 보고서: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). 경쟁 제품 비교: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) — agentmemory 대 mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**로컬 재현 방법:** [`eval/README.md`](../eval/README.md), LongMemEval `_s`(공개 500-Q)와 `coding-agent-life-v1`(자체 15-세션 코퍼스)을 위한 어댑터 플러그형 하니스입니다. grep / vector / agentmemory 어댑터를 나란히 평가하고, NDJSON으로 출력하며, 게시된 스코어카드는 [`docs/benchmarks/`](../docs/benchmarks/)에 보관됩니다.
-
-**다음과 함께 사용하기 좋습니다: [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), [Graphify](https://github.com/safishamsi/graphify).** 코드 그래프 인덱싱, 멀티 에이전트 빌드 파이프라인, 그리고 docs/PDF/이미지/비디오에 걸친 더 넓은 지식 그래프. agentmemory는 작업을 기억하고, 이 세 프로젝트는 나머지 컨텍스트 레이어를 밝혀줍니다. 레시피와 질문 라우팅 표: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="경쟁 제품 비교" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>내장 메모리 (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>유형</strong></td>
-<td>메모리 엔진 + MCP 서버</td>
-<td>메모리 레이어 API</td>
-<td>완전한 에이전트 런타임</td>
-<td>개인 AI</td>
-<td>메모리 API + 앱</td>
-<td>팀 메모리 허브 (LLM 프록시)</td>
-<td>벡터 메모리 (OSS)</td>
-<td>메모리 엔진 (Oracle DB)</td>
-<td>메모리 시스템</td>
-<td>정적 파일</td>
-</tr>
-<tr>
-<td><strong>검색 R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>해당 없음</td>
-<td>자체 보고</td>
-<td>PersonaMem 76% (자체 보고)</td>
-<td>~96.6% (자체 보고)</td>
-<td>94.4% (자체 보고)</td>
-<td>해당 없음</td>
-<td>해당 없음 (grep)</td>
-</tr>
-<tr>
-<td><strong>자동 캡처</strong></td>
-<td>hook 12개 (수동 작업 없음)</td>
-<td>수동 <code>add()</code> 호출</td>
-<td>에이전트 자체 편집</td>
-<td>수동</td>
-<td>API 측 추출</td>
-<td>프록시 가로채기 (base-URL 교체)</td>
-<td>수동</td>
-<td>API 추출</td>
-<td>수동</td>
-<td>수동 편집</td>
-</tr>
-<tr>
-<td><strong>검색</strong></td>
-<td>BM25 + Vector + Graph (RRF 융합)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>시맨틱</td>
-<td>Vector + RAG</td>
-<td>4가지 자산 유형 (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vector 전용</td>
-<td>Vector + 시맨틱</td>
-<td>감쇠 가중</td>
-<td>모든 것을 컨텍스트에 로드</td>
-</tr>
-<tr>
-<td><strong>멀티 에이전트</strong></td>
-<td>MCP + REST + lease + signal</td>
-<td>API (조정 없음)</td>
-<td>Letta 런타임 내에서만</td>
-<td>없음</td>
-<td>없음</td>
-<td>팀 역할 + 공유 자산</td>
-<td>없음</td>
-<td>스코프만 지원</td>
-<td>멀티 에이전트 공유</td>
-<td>에이전트별 파일</td>
-</tr>
-<tr>
-<td><strong>프레임워크 종속성</strong></td>
-<td>없음 (모든 MCP 클라이언트)</td>
-<td>없음</td>
-<td>높음 (Letta 사용 필수)</td>
-<td>독립형</td>
-<td>없음</td>
-<td>프록시가 모든 모델 호출을 프론팅</td>
-<td>없음</td>
-<td>Oracle Database</td>
-<td>없음</td>
-<td>에이전트별 포맷</td>
-</tr>
-<tr>
-<td><strong>외부 의존성</strong></td>
-<td>별도 데이터베이스 서비스 불필요</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + 벡터 DB</td>
-<td>다수</td>
-<td>매니지드 클라우드</td>
-<td>Docker 스택 (Core + Hub + Proxy)</td>
-<td>벡터 스토어</td>
-<td>Oracle AI Database</td>
-<td>없음</td>
-<td>없음</td>
-</tr>
-<tr>
-<td><strong>메모리 라이프사이클</strong></td>
-<td>4-tier 통합 + 감쇠 + 자동 망각</td>
-<td>수동적 추출</td>
-<td>에이전트 관리</td>
-<td>수동</td>
-<td>자동 망각</td>
-<td>수동 리뷰; 자동 라우팅 진행 중</td>
-<td>없음</td>
-<td>명시 없음</td>
-<td>감쇠 + 통합</td>
-<td>수동 정리</td>
-</tr>
-<tr>
-<td><strong>토큰 효율</strong></td>
-<td>세션당 ~1,900 토큰 ($10/년)</td>
-<td>통합 방식에 따라 다름</td>
-<td>핵심 메모리는 컨텍스트에 상주</td>
-<td>다양</td>
-<td>클라우드 가격 책정</td>
-<td>명시 없음</td>
-<td>토큰 예산 없음</td>
-<td>LLM 기반 (다양)</td>
-<td>다양</td>
-<td>관측 240개 기준 22K+ 토큰</td>
-</tr>
-<tr>
-<td><strong>실시간 뷰어</strong></td>
-<td>있음 (포트 3113)</td>
-<td>클라우드 대시보드</td>
-<td>클라우드 대시보드</td>
-<td>웹 UI</td>
-<td>클라우드 대시보드</td>
-<td>Hub 웹 UI</td>
-<td>없음</td>
-<td>없음</td>
-<td>없음</td>
-<td>없음</td>
-</tr>
-<tr>
-<td><strong>셀프 호스팅</strong></td>
-<td>예 (기본값)</td>
-<td>선택 사항</td>
-<td>선택 사항</td>
-<td>예</td>
-<td>아니오 (클라우드 전용)</td>
-<td>예 (Docker)</td>
-<td>예</td>
-<td>예 (Oracle DB)</td>
-<td>예</td>
-<td>예</td>
-</tr>
-</table>
-
-<sub>벤치마크 참고: agentmemory의 R@5만이 우리가 직접 측정한 결과입니다(LongMemEval-S, <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>에서 재현 가능). mem0와 Letta 수치는 그들이 게시한 LoCoMo 수치(다른 데이터셋)이며, MemPalace, supermemory, TencentDB (PersonaMem), oracleagentmemory 수치는 우리가 독립적으로 재현하지 않은 벤더 자체 보고 주장입니다(oracleagentmemory의 실행은 Oracle AI Database에 대해 GPT-5.5를 사용했습니다). 대략적인 비교를 위해 나란히 표시했을 뿐, 동일 데이터에 대한 정면 대결이 아닙니다. Star 수는 근사치이며 시간이 지나면서 변동합니다.</sub>
-
-알아둘 만한 **최근 진입자들**, [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md)에서 심층 비교:
-
-| 시스템 | ⭐ | 접근 방식 |
-|--------|---|-------|
-| Zep / Graphiti | 30K | 시간적 지식 그래프. 게시된 시간 쿼리 결과 중 가장 강력(LongMemEval 63.8%)하지만, 그래프가 비동기로 빌드되어 최신 사실이 지연될 수 있음 |
-| Cognee | 30K | 문서-지식 그래프 수집, Python 전용, 세션 캡처보다는 구조화된 엔티티 추출을 위해 설계됨 |
+출처를 포함한 최신 제품 및 배포 방식 비교:
-AgentMemory는 세션 캡처, 검색 가능한 프로젝트 이력, 로컬 뷰어를 함께 제공하며 처음부터 키 없이 사용할 수 있습니다.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -778,7 +570,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
메모리 서버 시작: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add`를 통한 네이티브 skill (50+ 에이전트)
+#### `npx skills add`를 통한 네이티브 skill
agentmemory는 Claude-Code 스타일의 `<dir>/SKILL.md` 형식으로 17개의 skill을 제공합니다: 호출 가능한 action skill 9개(`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`)와 에이전트가 필요할 때 로드하는 참조 skill 8개(`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`)입니다. 참조 skill은 소스에서 생성된 데이터 표를 담고 있어 절대 어긋나지 않습니다. vercel-labs의 [`skills`](https://npmjs.com/package/skills) CLI는 호출한 에이전트의 네이티브 skill 디렉터리에, 50+개의 에이전트(Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf 등)에 걸쳐 자동으로 설치합니다:
@@ -835,7 +627,7 @@ agentmemory 항목은 `mcpServers` 형태를 사용하는 모든 호스트(Curso
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks`는 공유 커스터마이징 디렉터리에 MCP와 캡처 hooks를 설치합니다. [Antigravity 설정 및 제한사항](../docs/plugins/antigravity.md) 참고. |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks`는 현재 IDE 버전과 동일한 MCP 및 hook 설정을 사용합니다. 기존 설치는 `--force`로 갱신해야 합니다. [업그레이드 노트](../docs/plugins/antigravity.md) 참고. |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro`는 사용자 수준 설정을 작성합니다. workspace 재정의는 코드 옆의 `.kiro/settings/mcp.json`에 둡니다. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp`는 표준 `mcpServers` 블록을 작성합니다. Warp는 `.claude/skills/`에서 skill도 자동으로 탐색합니다. Claude Code 플러그인을 설치하면 8개의 agentmemory skill(`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`)이 Warp의 슬래시 명령 팔레트에 네이티브로 나타납니다. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp`는 표준 `mcpServers` 블록을 작성합니다. Warp는 `.claude/skills/`에서 skill도 자동으로 탐색합니다. Claude Code 플러그인을 설치하면 8개의 agentmemory skill(`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`)이 Warp의 슬래시 명령 팔레트에 네이티브로 나타납니다. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline`은 표준 `mcpServers` 블록을 작성합니다. VS Code 확장 사용자는 Cline Settings → MCP Servers → Edit JSON을 통해 동일한 블록을 붙여넣으십시오. |
| **Continue.dev** | `~/.continue/config.yaml` (권장) 또는 `config.json` (레거시) | `agentmemory connect continue`는 설정이 없으면 `config.yaml`을 만들고 기존 `config.json`은 수정합니다. 기존 YAML은 주석과 앵커를 보존하기 위해 변경하지 않고 `mcpServers:` 아래에 붙일 블록을 출력합니다. Continue의 `mcpServers`는 배열입니다. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed`는 `context_servers`(Zed의 키이며 `mcpServers`가 아닙니다) 아래에 작성합니다. 원격 MCP 서버는 대신 `{"url": "..."}`로 연결할 수 있습니다. |
@@ -932,7 +724,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| 관리 | 수동으로 관리하는 지침 | 색인된 세션 이력과 메모리 |
| 검색 | 모든 것을 컨텍스트에 로드 | BM25 + vector + graph (top-K만) |
-| 토큰 비용 | 관측 240개 기준 22K+ | ~1,900 토큰 (92% 감소) |
| 에이전트 간 | 에이전트별 파일 | MCP + REST (모든 에이전트) |
| 조정 | 없음 | lease, signal, action, routine |
| 관측 가능성 | 파일을 수동으로 읽기 | :3113의 실시간 뷰어 |
@@ -945,7 +736,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1030,7 +821,7 @@ Reciprocal Rank Fusion(RRF, k=60)으로 융합되고, 세션별로 다변화됩
벡터 인덱스가 채워져 있으면, (`memory_recall` 뒤에 있는) `mem::search`는 하이브리드 BM25 + vector 랭커를 사용합니다. 임베딩이 없으면 BM25를 사용합니다. `smart-search`는 그래프 데이터가 존재하면 키 없는 모드에서도 구조적 그래프 매치를 추가로 융합할 수 있습니다. Lesson 리콜은 쿼리마다 전체 코퍼스를 스캔하는 대신 전용 인메모리 BM25 인덱스에서 동작합니다. 대체된 메모리 버전은 모든 리콜 경로에서 제외되며, 버전 체인이 그 기록을 보존합니다.
-벡터는 크래시나 강제 종료에도 유지됩니다. 벡터 인덱스는 최소 `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`(10분) 간격으로 버킷 단위로 저장됩니다. 그 사이에 추가되거나 제거된 모든 벡터는 상태 저장소의 작은 pending 로그에도 즉시 기록되며, 다음 시작 시 임베딩 프로바이더를 호출하지 않고 이를 재생합니다. 저장이 성공할 때마다 로그가 비워집니다. 재생 후에도 벡터가 없는 문서는 `AGENTMEMORY_VECTOR_BACKFILL_MAX`(500)개 단위 배치로 백그라운드에서 다시 임베딩되며, 더 이상 남지 않을 때까지 계속됩니다. 중단된 백필은 다음 시작 시 이어서 진행됩니다. `/agentmemory/status`와 뷰어는 pending 로그 크기와 백필 상태를 보여줍니다. 키 없는 설치는 아무것도 기록하지 않습니다.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25는 기본적으로 그리스어, 키릴 문자, 히브리어, 아랍어, 강세 부호가 있는 라틴 문자를 토크나이즈합니다. 중국어 / 일본어 / 한국어 메모리의 경우, 선택적 세그멘터(`npm install @node-rs/jieba tiny-segmenter`)를 설치해 CJK 런을 단어 수준 토큰으로 분할하십시오. 설치하지 않으면 agentmemory는 전체 런 토크나이제이션으로 soft fallback하고 stderr에 일회성 힌트를 출력합니다.
@@ -1046,9 +837,9 @@ EMBEDDING_PROVIDER=local
| 프로바이더 | 모델 | 비용 | 참고 |
|---|---|---|---|
-| **Local (권장 opt-in)** | `all-MiniLM-L6-v2` | 무료 | 첫 모델 다운로드 이후 온디바이스 실행, BM25 단독 대비 recall +8pp |
+| **Local (권장 opt-in)** | `all-MiniLM-L6-v2` | 무료 | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | 무료 티어 | 100+개 언어, 768/1536/3072차원(MRL), 2048 토큰 입력. `text-embedding-004`를 대체함([지원 종료, 2026년 1월 14일 중단](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | 최고 품질 |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | 유료 | 코드에 최적화됨 |
| Cohere | `embed-english-v3.0` | 무료 체험 | 범용 |
| OpenRouter | 모든 모델 | 다양함 | 멀티 모델 프록시 |
diff --git a/READMEs/README.nb-NO.md b/READMEs/README.nb-NO.md
index b8bc28c5..620048b4 100644
--- a/READMEs/README.nb-NO.md
+++ b/READMEs/README.nb-NO.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -329,223 +327,17 @@ Nyeste versjonsnotater: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Nøyaktighet ved gjenfinning
-
-**coding-agent-life-v1** (eget korpus, reproduserbart i sandbox)
-
-| Adapter | P@5 | R@5 | Topp-5-treffrate | p50-latens |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100 % topp-5-treffrate ved **P@5-matematikktaket** for dette korpuset (0.240, se scorekortet). Hybrid finner hver gullsesjon; grep mister 1 av 2 gull på det flersesjons-temporale søket. Gevinsten er **recall + temporal**, ikke samlet presisjon. Denne ytelsestesten er liten og gull-sparsom; den større LongMemEval-S nedenfor skiller bedre. Full oppdeling per type + korrigeringsnotat: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 spørsmål)
+Oppdaterte søkeresultater, definisjoner av måltall og trinn for reproduksjon:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Kun BM25-fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tokenbesparelser
-
-| Metode | Tokens/år | Kostnad/år |
-|---|---|---|
-| Lim inn full kontekst | 19.5M+ | Umulig (overskrider konteksten) |
-| LLM-oppsummert | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokale embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding-modell: `all-MiniLM-L6-v2` (lokal, gratis, ingen API-nøkkel). Fullstendige rapporter: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Konkurrentsammenligning: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) dekker agentmemory mot mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reprodusér lokalt:** [`eval/README.md`](../eval/README.md), en adapter-plugbar harness for LongMemEval `_s` (offentlig 500 spørsmål) + `coding-agent-life-v1` (eget korpus på 15 sesjoner). Grep-, vektor- og agentmemory-adaptere scores side ved side, NDJSON-utdata, publiserte scorekort havner i [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Fungerer godt sammen med [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), og [Graphify](https://github.com/safishamsi/graphify).** Kodegrafindeksering, flerstegs byggepipeliner, og bredere kunnskapsgrafer over dokumenter/PDF-er/bilder/videoer. agentmemory husker arbeidet; disse tre prosjektene lyser opp resten av kontekstlaget. Oppskrifter + spørsmål-rutingstabell: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Innebygd (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Type</strong></td>
-<td>Minnemotor + MCP-server</td>
-<td>Minnelag-API</td>
-<td>Fullstendig agent-runtime</td>
-<td>Personlig AI</td>
-<td>Minne-API + app</td>
-<td>Team-minnehub (LLM-proxy)</td>
-<td>Vektorminne (åpen kildekode)</td>
-<td>Minnemotor (Oracle DB)</td>
-<td>Minnesystem</td>
-<td>Statisk fil</td>
-</tr>
-<tr>
-<td><strong>Gjenfinning R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>Ikke tilgjengelig</td>
-<td>Selvrapportert</td>
-<td>PersonaMem 76% (selvrapportert)</td>
-<td>~96.6% (selvrapportert)</td>
-<td>94.4% (selvrapportert)</td>
-<td>Ikke tilgjengelig</td>
-<td>Ikke tilgjengelig (grep)</td>
-</tr>
-<tr>
-<td><strong>Automatisk fangst</strong></td>
-<td>12 hooks (ingen manuell innsats)</td>
-<td>Manuelle <code>add()</code>-kall</td>
-<td>Agenten redigerer selv</td>
-<td>Manuelt</td>
-<td>Uttrekk på API-siden</td>
-<td>Proxy-avlytting (base-URL-bytte)</td>
-<td>Manuelt</td>
-<td>API-uttrekk</td>
-<td>Manuelt</td>
-<td>Manuell redigering</td>
-</tr>
-<tr>
-<td><strong>Søk</strong></td>
-<td>BM25 + vektor + graf (RRF-fusjon)</td>
-<td>Vektor + graf</td>
-<td>Vektor (arkiv)</td>
-<td>Semantisk</td>
-<td>Vektor + RAG</td>
-<td>4 ressurstyper (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Kun vektor</td>
-<td>Vektor + semantisk</td>
-<td>Forfallsvektet</td>
-<td>Laster alt inn i konteksten</td>
-</tr>
-<tr>
-<td><strong>Flere agenter</strong></td>
-<td>MCP + REST + leaser + signaler</td>
-<td>API (ingen koordinering)</td>
-<td>Kun innenfor Letta-runtime</td>
-<td>Nei</td>
-<td>Nei</td>
-<td>Teamroller + delte ressurser</td>
-<td>Nei</td>
-<td>Kun avgrenset</td>
-<td>Delt mellom flere agenter</td>
-<td>Filer per agent</td>
-</tr>
-<tr>
-<td><strong>Rammeverksbinding</strong></td>
-<td>Ingen (enhver MCP-klient)</td>
-<td>Ingen</td>
-<td>Høy (må bruke Letta)</td>
-<td>Frittstående</td>
-<td>Ingen</td>
-<td>Proxy fronter hvert modellkall</td>
-<td>Ingen</td>
-<td>Oracle Database</td>
-<td>Ingen</td>
-<td>Format per agent</td>
-</tr>
-<tr>
-<td><strong>Eksterne avhengigheter</strong></td>
-<td>Ingen separat databasetjeneste nødvendig</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektor-DB</td>
-<td>Flere</td>
-<td>Administrert sky</td>
-<td>Docker-stack (Core + Hub + Proxy)</td>
-<td>Vektorlager</td>
-<td>Oracle AI Database</td>
-<td>Ingen</td>
-<td>Ingen</td>
-</tr>
-<tr>
-<td><strong>Minnelivssyklus</strong></td>
-<td>4-nivås konsolidering + forfall + automatisk glemming</td>
-<td>Passivt uttrekk</td>
-<td>Agent-administrert</td>
-<td>Manuelt</td>
-<td>Automatisk glemming</td>
-<td>Manuell gjennomgang; automatisk ruting underveis</td>
-<td>Ingen</td>
-<td>Ikke angitt</td>
-<td>Forfall + konsolidering</td>
-<td>Manuell beskjæring</td>
-</tr>
-<tr>
-<td><strong>Tokeneffektivitet</strong></td>
-<td>~1 900 tokens/sesjon ($10/år)</td>
-<td>Varierer med integrasjon</td>
-<td>Kjerneminne i konteksten</td>
-<td>Varierer</td>
-<td>Skyprising</td>
-<td>Ikke angitt</td>
-<td>Ingen tokenbudsjett</td>
-<td>LLM-basert (varierer)</td>
-<td>Varierer</td>
-<td>22K+ tokens ved 240 observasjoner</td>
-</tr>
-<tr>
-<td><strong>Viewer i realtid</strong></td>
-<td>Ja (port 3113)</td>
-<td>Sky-dashbord</td>
-<td>Sky-dashbord</td>
-<td>Web-UI</td>
-<td>Sky-dashbord</td>
-<td>Hub-web-UI</td>
-<td>Nei</td>
-<td>Nei</td>
-<td>Nei</td>
-<td>Nei</td>
-</tr>
-<tr>
-<td><strong>Selvhostet</strong></td>
-<td>Ja (standard)</td>
-<td>Valgfritt</td>
-<td>Valgfritt</td>
-<td>Ja</td>
-<td>Nei (kun sky)</td>
-<td>Ja (Docker)</td>
-<td>Ja</td>
-<td>Ja (Oracle DB)</td>
-<td>Ja</td>
-<td>Ja</td>
-</tr>
-</table>
-
-<sub>Merk om ytelsestester: kun agentmemorys R@5 er vårt eget målte resultat (LongMemEval-S, reproduserbart fra <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Tallene for mem0 og Letta er deres publiserte LoCoMo-tall (et annet datasett); tallene for MemPalace, supermemory, TencentDB (PersonaMem) og oracleagentmemory er leverandørers selvrapporterte påstander vi ikke har reprodusert selv (oracleagentmemorys kjøring brukte GPT-5.5 mot en Oracle AI Database). Vist side ved side kun som et grovt sammenligningsgrunnlag, ikke en direkte sammenligning på identiske data. Stjernetall er tilnærmede og endrer seg over tid.</sub>
-
-**Nyere aktører** verdt å kjenne til, sammenlignet i detalj i [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | Vinkling |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal kunnskapsgraf; de sterkeste publiserte resultatene på temporale søk (LongMemEval 63.8%), men grafen bygges asynkront så ferske fakta kan ligge etter |
-| Cognee | 30K | Dokument-til-kunnskapsgraf-innlasting, kun Python, bygget for strukturert entitetsuttrekk snarere enn sesjonsfangst |
+Oppdatert sammenligning av produkter og driftsalternativer med kilder:
-AgentMemory kombinerer registrering fra kodeagenter, søkbar prosjekthistorikk og en lokal viewer. Du kan arbeide uten nøkkel fra første start.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -777,7 +569,7 @@ Full veiledning: [`integrations/hermes/`](../integrations/hermes/)
Start minneserveren: `npx -y @agentmemory/agentmemory@latest`
-#### Native skills via `npx skills add` (50+ agenter)
+#### Native skills via `npx skills add`
agentmemory leveres med 17 skills i Claude-Code-stilens `<dir>/SKILL.md`-format: 9 skills som kan kalles (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) og 8 referanseskills agenten laster ved behov (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Referanseskillene har datatabeller generert fra kildekoden, så de aldri går ut av synk. CLI-en [`skills`](https://npmjs.com/package/skills) fra vercel-labs installerer dem automatisk inn i den kallende agentens native skill-katalog over 50+ agenter (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, med flere):
@@ -834,7 +626,7 @@ agentmemory-oppføringen er **samme MCP-server-blokk** på tvers av alle verter
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installerer MCP og fangst-hooks i den delte tilpasningskatalogen. Se [Antigravity-oppsett og begrensninger](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` bruker samme MCP- og hook-konfigurasjon som nåværende IDE-versjoner. Eksisterende installasjoner bør oppdateres med `--force`; se [oppgraderingsnotatene](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` skriver brukernivå-konfigurasjonen. Arbeidsområde-overstyringer går i `.kiro/settings/mcp.json` ved siden av koden din. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver standard `mcpServers`-blokk. Warp oppdager også automatisk skills fra `.claude/skills/`; når Claude Code-pluginen er installert vises de 8 agentmemory-skillene (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) naturlig i Warps slash-kommando-palett. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver standard `mcpServers`-blokk. Warp oppdager også automatisk skills fra `.claude/skills/`; når Claude Code-pluginen er installert vises de 9 agentmemory-skillene (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) naturlig i Warps slash-kommando-palett. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` skriver standard `mcpServers`-blokk. VS Code-utvidelsesbrukere: lim inn samme blokk via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (foretrukket) eller `config.json` (eldre) | `agentmemory connect continue` oppretter en manglende `config.yaml` eller endrer eksisterende `config.json`. Hvis YAML-filen finnes, vises blokken for `mcpServers:` slik at kommentarer og ankere bevares. Continue bruker et array, ikke et objekt. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` skriver under `context_servers` (Zeds nøkkel, IKKE `mcpServers`). Eksterne MCP-servere kan kobles til via `{"url": "..."}` i stedet. |
@@ -931,7 +723,6 @@ Hver AI-kodeagent leveres med innebygd hukommelse: Claude Code har `MEMORY.md`,
|---|---|---|
| Vedlikehold | Manuelt vedlikeholdte instruksjoner | Indeksert økthistorikk og minner |
| Søk | Laster alt inn i konteksten | BM25 + vektor + graf (kun topp-K) |
-| Tokenkostnad | 22K+ ved 240 observasjoner | ~1 900 tokens (92 % mindre) |
| Tvers av agenter | Filer per agent | MCP + REST (hvilken som helst agent) |
| Koordinering | Ingen | Leaser, signaler, handlinger, rutiner |
| Observerbarhet | Les filer manuelt | Viewer i realtid på :3113 |
@@ -944,7 +735,7 @@ Hver AI-kodeagent leveres med innebygd hukommelse: Claude Code har `MEMORY.md`,
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1029,7 +820,7 @@ Fusjonert med Reciprocal Rank Fusion (RRF, k=60) og sesjonsdiversifisert (maks 3
Når en vektorindeks er fylt, bruker `mem::search` (bak `memory_recall`) den hybride BM25 + vektor-rangereren. Uten embeddings bruker den BM25. `smart-search` kan i tillegg slå sammen strukturelle grafmatcher når grafdata finnes, også i nøkkelfri modus. Gjenkalling av læringspunkter kjører på en dedikert BM25-indeks i minnet i stedet for å skanne hele korpuset per søk. Erstattede minneversjoner er utelatt fra hver gjenkallingsvei; versjonskjeden beholder historikken deres.
-Vektorer overlever en krasj eller tvungen avslutning. Vektorindeksen lagres i bøtter minst hvert `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutter). Hver vektor som legges til eller fjernes i mellomtiden skrives også umiddelbart til en liten ventende logg i state-lageret, og neste oppstart spiller den av igjen uten å kalle embedding-leverandøren. Hver vellykkede lagring tømmer loggen. Dokumenter som fortsatt ikke har en vektor etter avspillingen, re-embeddes i bakgrunnen i batcher av `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) til ingen er igjen, og en etterfylling som stoppes fortsetter ved neste oppstart. `/agentmemory/status` og viewer viser størrelsen på den ventende loggen og etterfyllingsstatusen. Nøkkelfrie installasjoner skriver ingenting.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokeniserer gresk, kyrillisk, hebraisk, arabisk, og aksentuert latin ut av boksen. For kinesiske/japanske/koreanske minner, installer de valgfrie segmenterne (`npm install @node-rs/jieba tiny-segmenter`) for å dele opp CJK-sekvenser i ord-nivå-tokens; uten dem faller agentmemory mykt tilbake til hele-sekvens-tokenisering og skriver et engangs-hint til stderr.
@@ -1045,9 +836,9 @@ Den vanlige npm-installasjonen inkluderer den valgfrie `@huggingface/transformer
| Leverandør | Modell | Kostnad | Merknader |
|---|---|---|---|
-| **Lokal (anbefalt opt-in)** | `all-MiniLM-L6-v2` | Gratis | På enheten etter den første modellnedlastingen, +8 prosentpoeng gjenkalling over kun-BM25 |
+| **Lokal (anbefalt opt-in)** | `all-MiniLM-L6-v2` | Gratis | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Gratis nivå | 100+ språk, 768/1536/3072 dimensjoner (MRL), 2048-tokens inndata. Erstatter `text-embedding-004` ([avviklet, stengt 14. januar 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Høyest kvalitet |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Betalt | Optimalisert for kode |
| Cohere | `embed-english-v3.0` | Gratis prøveperiode | Generell bruk |
| OpenRouter | Hvilken som helst modell | Varierer | Flermodell-proxy |
diff --git a/READMEs/README.nl-NL.md b/READMEs/README.nl-NL.md
index 864d483c..835d336c 100644
--- a/READMEs/README.nl-NL.md
+++ b/READMEs/README.nl-NL.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -329,223 +327,17 @@ Laatste release notes: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval-nauwkeurigheid
-
-**coding-agent-life-v1** (eigen corpus, reproduceerbaar in een sandbox)
-
-| Adapter | P@5 | R@5 | Top-5-hitrate | p50-latentie |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% top-5-hitrate op het **P@5-wiskundige plafond** voor dit corpus (0.240, zie scorecard). Hybride haalt elke gold session op; grep mist 1 van 2 gold-resultaten bij de multi-session temporele query. De winst zit in **recall + temporeel**, niet in geaggregeerde precisie. Deze benchmark is klein en gold-sparse; de grotere LongMemEval-S hieronder onderscheidt beter. Volledige uitsplitsing per type + correctienotitie: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 vragen)
+Actuele zoekresultaten, definities van meetwaarden en stappen om de test te herhalen:
-| Systeem | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tokenbesparing
-
-| Aanpak | Tokens/jaar | Kosten/jaar |
-|---|---|---|
-| Volledige context plakken | 19.5M+ | Onmogelijk (overschrijdt het venster) |
-| LLM-samengevat | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokale embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embeddingmodel: `all-MiniLM-L6-v2` (lokaal, gratis, geen API-key nodig). Volledige rapporten: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Vergelijking met concurrenten: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), met agentmemory tegenover mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee en Hippo.
-
-**Lokaal reproduceren:** [`eval/README.md`](../eval/README.md), een harness met inpluggbare adapters voor LongMemEval `_s` (publieke set van 500 vragen) + `coding-agent-life-v1` (eigen corpus van 15 sessies). Grep-, vector- en agentmemory-adapters scoren naast elkaar, NDJSON-output, gepubliceerde scorecards komen terecht in [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Combineert goed met [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) en [Graphify](https://github.com/safishamsi/graphify).** Code-graph-indexering, multi-agent build-pipelines en bredere knowledge graphs over docs / PDF's / afbeeldingen / video's. agentmemory onthoudt het werk; die drie projecten verlichten de rest van de context-laag. Recepten + tabel voor vraagroutering: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Concurrenten" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Ingebouwd (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Type</strong></td>
-<td>Memory engine + MCP-server</td>
-<td>Memory-laag-API</td>
-<td>Volledige agent-runtime</td>
-<td>Persoonlijke AI</td>
-<td>Memory-API + app</td>
-<td>Team-memory-hub (LLM-proxy)</td>
-<td>Vector-memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Memory-systeem</td>
-<td>Statisch bestand</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Zelf gerapporteerd</td>
-<td>PersonaMem 76% (zelf gerapporteerd)</td>
-<td>~96.6% (zelf gerapporteerd)</td>
-<td>94.4% (zelf gerapporteerd)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hooks (geen handmatige moeite)</td>
-<td>Handmatige <code>add()</code>-aanroepen</td>
-<td>Agent bewerkt zelf</td>
-<td>Handmatig</td>
-<td>Extractie aan de API-kant</td>
-<td>Proxy-interceptie (base-URL wisselen)</td>
-<td>Handmatig</td>
-<td>API-extractie</td>
-<td>Handmatig</td>
-<td>Handmatig bewerken</td>
-</tr>
-<tr>
-<td><strong>Zoeken</strong></td>
-<td>BM25 + Vector + Graph (RRF-fusie)</td>
-<td>Vector + Graph</td>
-<td>Vector (archief)</td>
-<td>Semantisch</td>
-<td>Vector + RAG</td>
-<td>4 asset-types (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Alleen vector</td>
-<td>Vector + semantisch</td>
-<td>Decay-gewogen</td>
-<td>Laadt alles in de context</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (geen coördinatie)</td>
-<td>Alleen binnen Letta-runtime</td>
-<td>Nee</td>
-<td>Nee</td>
-<td>Teamrollen + gedeelde assets</td>
-<td>Nee</td>
-<td>Alleen scoped</td>
-<td>Gedeeld tussen meerdere agents</td>
-<td>Bestanden per agent</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>Geen (elke MCP-client)</td>
-<td>Geen</td>
-<td>Hoog (moet Letta gebruiken)</td>
-<td>Standalone</td>
-<td>Geen</td>
-<td>Proxy staat voor elke model-aanroep</td>
-<td>Geen</td>
-<td>Oracle Database</td>
-<td>Geen</td>
-<td>Formaat per agent</td>
-</tr>
-<tr>
-<td><strong>Externe dependencies</strong></td>
-<td>Geen afzonderlijke databaseservice nodig</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector-db</td>
-<td>Meerdere</td>
-<td>Managed cloud</td>
-<td>Docker-stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Geen</td>
-<td>Geen</td>
-</tr>
-<tr>
-<td><strong>Memory lifecycle</strong></td>
-<td>4-laags consolidatie + decay + auto-forget</td>
-<td>Passieve extractie</td>
-<td>Beheerd door agent</td>
-<td>Handmatig</td>
-<td>Auto-forget</td>
-<td>Handmatige review; auto-routing in ontwikkeling</td>
-<td>Geen</td>
-<td>Niet vermeld</td>
-<td>Decay + consolidatie</td>
-<td>Handmatig opschonen</td>
-</tr>
-<tr>
-<td><strong>Tokenefficiëntie</strong></td>
-<td>~1,900 tokens/sessie ($10/jaar)</td>
-<td>Varieert per integratie</td>
-<td>Core memory in context</td>
-<td>Varieert</td>
-<td>Cloud pricing</td>
-<td>Niet vermeld</td>
-<td>Geen tokenbudget</td>
-<td>LLM-ondersteund (varieert)</td>
-<td>Varieert</td>
-<td>22K+ tokens bij 240 observaties</td>
-</tr>
-<tr>
-<td><strong>Realtime viewer</strong></td>
-<td>Ja (poort 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>Nee</td>
-<td>Nee</td>
-<td>Nee</td>
-<td>Nee</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Ja (standaard)</td>
-<td>Optioneel</td>
-<td>Optioneel</td>
-<td>Ja</td>
-<td>Nee (alleen cloud)</td>
-<td>Ja (Docker)</td>
-<td>Ja</td>
-<td>Ja (Oracle DB)</td>
-<td>Ja</td>
-<td>Ja</td>
-</tr>
-</table>
-
-<sub>Opmerking over de benchmark: alleen de R@5 van agentmemory is ons eigen gemeten resultaat (LongMemEval-S, reproduceerbaar via <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). De cijfers van mem0 en Letta zijn hun gepubliceerde LoCoMo-cijfers (een andere dataset); de cijfers van MemPalace, supermemory, TencentDB (PersonaMem) en oracleagentmemory zijn door de leverancier zelf gerapporteerde claims die we niet onafhankelijk hebben gereproduceerd (de run van oracleagentmemory gebruikte GPT-5.5 tegen een Oracle AI Database). Naast elkaar getoond voor een ruwe inschatting, niet als head-to-head op identieke data. Stertallen zijn benaderend en veranderen in de tijd.</sub>
-
-**Nieuwere spelers** die het kennen waard zijn, in detail vergeleken in [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Systeem | ⭐ | Invalshoek |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporele knowledge graph; sterkste gepubliceerde resultaten voor temporele queries (LongMemEval 63.8%), maar de graph wordt asynchroon opgebouwd, waardoor verse feiten kunnen achterblijven |
-| Cognee | 30K | Document-naar-knowledge-graph-ingestie, alleen Python, gebouwd voor gestructureerde entiteitsextractie in plaats van sessie-capture |
+Bijgewerkte vergelijking van producten en implementatieopties, met bronnen:
-AgentMemory combineert vastlegging door programmeeragents, doorzoekbare projectgeschiedenis en een lokale viewer. Je kunt vanaf de eerste start zonder sleutel werken.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -775,7 +567,7 @@ Volledige handleiding: [`integrations/hermes/`](../integrations/hermes/)
Start de memory server: `npx -y @agentmemory/agentmemory@latest`
-#### Native skills via `npx skills add` (50+ agents)
+#### Native skills via `npx skills add`
agentmemory levert 17 skills in het Claude-Code-achtige `<dir>/SKILL.md`-formaat: 9 aanroepbare actie-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) en 8 referentieskills die de agent op aanvraag laadt (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). De referentieskills bevatten datatabellen die uit de bron gegenereerd zijn, dus ze lopen nooit uit de pas. De [`skills`](https://npmjs.com/package/skills)-CLI van vercel-labs installeert ze automatisch in de native skills-map van de aanroepende agent, verspreid over 50+ agents (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, en meer):
@@ -832,7 +624,7 @@ De agentmemory-entry is hetzelfde **MCP-serverblok** op elke host die de `mcpSer
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installeert MCP en capture-hooks in de gedeelde customization-map. Zie [Antigravity-setup en -beperkingen](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` gebruikt dezelfde MCP- en hookconfiguratie als de huidige IDE-versies. Bestaande installaties moeten verversen met `--force`; zie de [upgrade-notities](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` schrijft de config op gebruikersniveau. Workspace-overrides komen in `.kiro/settings/mcp.json` naast je code. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` schrijft het standaard `mcpServers`-blok. Warp ontdekt ook automatisch skills vanuit `.claude/skills/`; zodra de Claude Code-plugin is geïnstalleerd, verschijnen de 8 agentmemory-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) native in het slashcommando-palet van Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` schrijft het standaard `mcpServers`-blok. Warp ontdekt ook automatisch skills vanuit `.claude/skills/`; zodra de Claude Code-plugin is geïnstalleerd, verschijnen de 9 agentmemory-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) native in het slashcommando-palet van Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` schrijft het standaard `mcpServers`-blok. Gebruikers van de VS Code-extensie: plak hetzelfde blok via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (voorkeur) of `config.json` (legacy) | `agentmemory connect continue` maakt een ontbrekende `config.yaml` of wijzigt een bestaande `config.json`. Bij bestaande YAML toont de adapter het blok voor `mcpServers:` zodat comments en anchors behouden blijven. Continue gebruikt een array, geen object. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` schrijft onder `context_servers` (de key van Zed, NIET `mcpServers`). Remote MCP-servers kunnen in plaats daarvan gekoppeld worden via `{"url": "..."}`. |
@@ -929,7 +721,6 @@ Elke AI coding agent levert ingebouwd geheugen: Claude Code heeft `MEMORY.md`, C
|---|---|---|
| Onderhoud | Handmatig bijgehouden instructies | Geïndexeerde sessiegeschiedenis en herinneringen |
| Zoeken | Laadt alles in de context | BM25 + vector + graph (alleen top-K) |
-| Tokenkosten | 22K+ bij 240 observaties | ~1,900 tokens (92% minder) |
| Cross-agent | Bestanden per agent | MCP + REST (elke agent) |
| Coördinatie | Geen | Leases, signals, actions, routines |
| Observability | Bestanden handmatig lezen | Realtime viewer op :3113 |
@@ -942,7 +733,7 @@ Elke AI coding agent levert ingebouwd geheugen: Claude Code heeft `MEMORY.md`, C
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1027,7 +818,7 @@ Samengevoegd met Reciprocal Rank Fusion (RRF, k=60) en gediversifieerd per sessi
Wanneer een vector-index gevuld is, gebruikt `mem::search` (achter `memory_recall`) de hybride BM25 + vector-ranker. Zonder embeddings gebruikt het BM25. `smart-search` kan daarnaast structurele graph-matches samenvoegen wanneer er graph-data bestaat, ook in keyless-modus. Lesson-recall draait op een eigen in-memory BM25-index in plaats van het hele corpus per query te doorzoeken. Vervangen memory-versies worden uitgesloten van elk recall-pad; de versieketen behoudt hun geschiedenis.
-Vectoren overleven een crash of force-kill. De vector-index wordt in batches opgeslagen, maximaal elke `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minuten). Elke vector die er tussendoor bijkomt of verdwijnt, wordt ook meteen geschreven naar een klein pending-log in de state store, en de volgende start speelt dit af zonder de embedding-provider aan te roepen. Elke succesvolle save maakt het log leeg. Documenten die na het afspelen nog geen vector hebben, worden op de achtergrond opnieuw ge-embed in batches van `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) totdat er geen meer over zijn, en een backfill die gestopt wordt, gaat bij de volgende start verder. `/agentmemory/status` en de viewer tonen de grootte van het pending-log en de status van de backfill. Keyless-installaties schrijven niets.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokeniseert standaard Grieks, Cyrillisch, Hebreeuws, Arabisch en Latijn met accenten. Voor Chinese / Japanse / Koreaanse memories installeer je de optionele segmenters (`npm install @node-rs/jieba tiny-segmenter`) om CJK-reeksen op te splitsen in tokens op woordniveau; zonder deze valt agentmemory soft terug op tokenisatie van de hele reeks en print het eenmalig een hint naar stderr.
@@ -1043,9 +834,9 @@ De normale npm-installatie bevat de optionele `@huggingface/transformers`-runtim
| Provider | Model | Kosten | Opmerkingen |
|---|---|---|---|
-| **Lokaal (aanbevolen opt-in)** | `all-MiniLM-L6-v2` | Gratis | Op het eigen apparaat na de eerste modeldownload, +8pp recall t.o.v. alleen BM25 |
+| **Lokaal (aanbevolen opt-in)** | `all-MiniLM-L6-v2` | Gratis | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Gratis tier | 100+ talen, 768/1536/3072 dims (MRL), input tot 2048 tokens. Vervangt `text-embedding-004` ([deprecated, uitgefaseerd op 14 jan. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Hoogste kwaliteit |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Betaald | Geoptimaliseerd voor code |
| Cohere | `embed-english-v3.0` | Gratis trial | Algemeen inzetbaar |
| OpenRouter | Elk model | Varieert | Multi-model-proxy |
diff --git a/READMEs/README.pl-PL.md b/READMEs/README.pl-PL.md
index 5bc50860..b1986723 100644
--- a/READMEs/README.pl-PL.md
+++ b/READMEs/README.pl-PL.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% trafności wyszukiwania R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% mniej tokenów" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 narzędzia MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automatycznych hooków" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 zewnętrznych baz danych" height="38" /></picture>
@@ -331,223 +329,17 @@ Najnowsze informacje o wydaniach: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarki" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Dokładność wyszukiwania
-
-**coding-agent-life-v1** (korpus własny, odtwarzalny w sandboksie)
-
-| Adapter | P@5 | R@5 | Trafienia w top-5 | opóźnienie p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% trafień w top-5 przy **matematycznym suficie P@5** dla tego korpusu (0.240, zobacz scorecard). Hybrid odzyskuje każdą złotą sesję; grep pomija 1 z 2 złotych sesji w zapytaniu temporalnym obejmującym wiele sesji. Przewaga dotyczy **recall + czasu**, a nie zagregowanej precyzji. Ten benchmark jest mały i ma mało danych złotych; większy LongMemEval-S poniżej różnicuje lepiej. Pełny podział per typ + uwaga o korekcie: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 pytań)
+Aktualne wyniki wyszukiwania, definicje metryk i kroki odtworzenia testu:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Fallback tylko BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Oszczędność tokenów
-
-| Podejście | Tokeny/rok | Koszt/rok |
-|---|---|---|
-| Wklejenie pełnego kontekstu | 19.5M+ | Niemożliwe (przekracza okno kontekstu) |
-| Podsumowane przez LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokalne embeddingi | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Model embeddingów: `all-MiniLM-L6-v2` (lokalny, bezpłatny, bez klucza API). Pełne raporty: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Porównanie z konkurencją: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) obejmujące agentmemory w porównaniu do mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Odtwórz lokalnie:** [`eval/README.md`](../eval/README.md), framework testowy z podłączanymi adapterami dla LongMemEval `_s` (publiczne 500 pytań) + `coding-agent-life-v1` (własny korpus 15 sesji). Adaptery grep / wektorowy / agentmemory punktowane obok siebie, wyjście w formacie NDJSON, opublikowane scorecardy trafiają do [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Współpracuje z [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) i [Graphify](https://github.com/safishamsi/graphify).** Indeksowanie grafu kodu, potoki budowania z wieloma agentami i szersze grafy wiedzy obejmujące dokumenty / PDF-y / obrazy / wideo. agentmemory pamięta pracę; te trzy projekty ożywiają resztę warstwy kontekstu. Przepisy + tabela routingu zapytań: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Porównanie z konkurencją" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Wbudowana (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Typ</strong></td>
-<td>Silnik pamięci + serwer MCP</td>
-<td>API warstwy pamięci</td>
-<td>Pełny runtime agenta</td>
-<td>Osobiste AI</td>
-<td>API pamięci + aplikacja</td>
-<td>Hub pamięci zespołowej (proxy LLM)</td>
-<td>Pamięć wektorowa (OSS)</td>
-<td>Silnik pamięci (Oracle DB)</td>
-<td>System pamięci</td>
-<td>Statyczny plik</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Dane producenta</td>
-<td>PersonaMem 76% (dane producenta)</td>
-<td>~96.6% (dane producenta)</td>
-<td>94.4% (dane producenta)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-przechwytywanie</strong></td>
-<td>12 hooków (zero pracy ręcznej)</td>
-<td>Ręczne wywołania <code>add()</code></td>
-<td>Agent edytuje się sam</td>
-<td>Ręczne</td>
-<td>Ekstrakcja po stronie API</td>
-<td>Przechwytywanie przez proxy (zamiana base-URL)</td>
-<td>Ręczne</td>
-<td>Ekstrakcja przez API</td>
-<td>Ręczne</td>
-<td>Ręczna edycja</td>
-</tr>
-<tr>
-<td><strong>Wyszukiwanie</strong></td>
-<td>BM25 + wektory + graf (fuzja RRF)</td>
-<td>Wektory + graf</td>
-<td>Wektory (archiwalne)</td>
-<td>Semantyczne</td>
-<td>Wektory + RAG</td>
-<td>4 typy zasobów (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Tylko wektory</td>
-<td>Wektory + semantyczne</td>
-<td>Ważone zanikaniem</td>
-<td>Wczytuje wszystko do kontekstu</td>
-</tr>
-<tr>
-<td><strong>Wiele agentów</strong></td>
-<td>MCP + REST + leasy + sygnały</td>
-<td>API (brak koordynacji)</td>
-<td>Tylko w ramach runtime Letta</td>
-<td>Nie</td>
-<td>Nie</td>
-<td>Role zespołowe + współdzielone zasoby</td>
-<td>Nie</td>
-<td>Tylko w ograniczonym zakresie</td>
-<td>Współdzielona między agentami</td>
-<td>Pliki per agent</td>
-</tr>
-<tr>
-<td><strong>Uzależnienie od frameworka</strong></td>
-<td>Brak (każdy klient MCP)</td>
-<td>Brak</td>
-<td>Wysokie (musisz używać Letta)</td>
-<td>Samodzielny</td>
-<td>Brak</td>
-<td>Proxy przechwytuje każde wywołanie modelu</td>
-<td>Brak</td>
-<td>Oracle Database</td>
-<td>Brak</td>
-<td>Format per agent</td>
-</tr>
-<tr>
-<td><strong>Zależności zewnętrzne</strong></td>
-<td>Nie wymaga osobnej usługi bazy danych</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + baza wektorowa</td>
-<td>Wiele</td>
-<td>Zarządzana chmura</td>
-<td>Stos Docker (Core + Hub + Proxy)</td>
-<td>Magazyn wektorowy</td>
-<td>Oracle AI Database</td>
-<td>Brak</td>
-<td>Brak</td>
-</tr>
-<tr>
-<td><strong>Cykl życia pamięci</strong></td>
-<td>4-poziomowa konsolidacja + zanikanie + auto-zapominanie</td>
-<td>Pasywna ekstrakcja</td>
-<td>Zarządzane przez agenta</td>
-<td>Ręczne</td>
-<td>Auto-zapominanie</td>
-<td>Ręczny przegląd; auto-routing w budowie</td>
-<td>Brak</td>
-<td>Nie podano</td>
-<td>Zanikanie + konsolidacja</td>
-<td>Ręczne czyszczenie</td>
-</tr>
-<tr>
-<td><strong>Efektywność tokenowa</strong></td>
-<td>~1,900 tokenów/sesję ($10/rok)</td>
-<td>Zależy od integracji</td>
-<td>Pamięć core w kontekście</td>
-<td>Zależy</td>
-<td>Cennik chmury</td>
-<td>Nie podano</td>
-<td>Brak budżetu tokenów</td>
-<td>Oparte na LLM (zależy)</td>
-<td>Zależy</td>
-<td>22K+ tokenów przy 240 obs.</td>
-</tr>
-<tr>
-<td><strong>Podgląd w czasie rzeczywistym</strong></td>
-<td>Tak (port 3113)</td>
-<td>Dashboard w chmurze</td>
-<td>Dashboard w chmurze</td>
-<td>Interfejs webowy</td>
-<td>Dashboard w chmurze</td>
-<td>Interfejs webowy Hub</td>
-<td>Nie</td>
-<td>Nie</td>
-<td>Nie</td>
-<td>Nie</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Tak (domyślnie)</td>
-<td>Opcjonalnie</td>
-<td>Opcjonalnie</td>
-<td>Tak</td>
-<td>Nie (tylko chmura)</td>
-<td>Tak (Docker)</td>
-<td>Tak</td>
-<td>Tak (Oracle DB)</td>
-<td>Tak</td>
-<td>Tak</td>
-</tr>
-</table>
-
-<sub>Uwaga dotycząca benchmarku: tylko wynik R@5 agentmemory jest naszym własnym zmierzonym wynikiem (LongMemEval-S, odtwarzalny na podstawie <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Wartości mem0 i Letta to ich opublikowane liczby LoCoMo (inny zbiór danych); wartości MemPalace, supermemory, TencentDB (PersonaMem) i oracleagentmemory to deklaracje producentów, których nie odtworzyliśmy niezależnie (uruchomienie oracleagentmemory używało GPT-5.5 względem Oracle AI Database). Zestawione obok siebie tylko orientacyjnie, nie jako test głowa w głowę na identycznych danych. Liczby gwiazdek są przybliżone i zmieniają się w czasie.</sub>
-
-**Nowsi uczestnicy** warci uwagi, porównani szerzej w [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | Podejście |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporalny graf wiedzy; najsilniejsze opublikowane wyniki zapytań temporalnych (LongMemEval 63.8%), ale graf buduje się asynchronicznie, więc świeże fakty mogą się spóźniać |
-| Cognee | 30K | Ingestia dokumentów do grafu wiedzy, tylko Python, zbudowane do strukturalnej ekstrakcji encji, a nie przechwytywania sesji |
+Aktualne porównanie produktów i opcji wdrożenia wraz ze źródłami:
-AgentMemory łączy przechwytywanie sesji, przeszukiwalną historię projektu i lokalny podgląd, z trybem bez klucza od pierwszego uruchomienia.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -784,7 +576,7 @@ Pełny przewodnik: [`integrations/hermes/`](../integrations/hermes/)
Uruchom serwer pamięci: `npx -y @agentmemory/agentmemory@latest`
-#### Natywne skille przez `npx skills add` (50+ agentów)
+#### Natywne skille przez `npx skills add`
agentmemory dostarcza 17 skilli w formacie `<dir>/SKILL.md` w stylu Claude Code: 9 wywoływalnych skilli akcji (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) oraz 8 skilli referencyjnych wczytywanych przez agenta na żądanie (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Skille referencyjne zawierają tabele danych generowane ze źródła, więc nigdy się nie rozjeżdżają. CLI [`skills`](https://npmjs.com/package/skills) od vercel-labs automatycznie instaluje je w natywnym katalogu skilli wywołującego agenta, obejmując ponad 50 agentów (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf i inne):
@@ -842,7 +634,7 @@ Wpis agentmemory to **ten sam blok serwera MCP** we wszystkich hostach, które u
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` instaluje MCP i hooki przechwytywania we współdzielonym katalogu personalizacji. Zobacz [Konfiguracja i ograniczenia Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` używa tej samej konfiguracji MCP i hooków, jak aktualne wersje IDE. Istniejące instalacje powinny odświeżyć się za pomocą `--force`; zobacz [uwagi dotyczące aktualizacji](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` zapisuje konfigurację na poziomie użytkownika. Nadpisania na poziomie workspace idą do `.kiro/settings/mcp.json` obok twojego kodu. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` zapisuje standardowy blok `mcpServers`. Warp automatycznie wykrywa też skille z `.claude/skills/`; gdy zainstalowany jest plugin Claude Code, 8 skilli agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) pojawia się natywnie w palecie komend ukośnikowych Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` zapisuje standardowy blok `mcpServers`. Warp automatycznie wykrywa też skille z `.claude/skills/`; gdy zainstalowany jest plugin Claude Code, 9 skilli agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) pojawia się natywnie w palecie komend ukośnikowych Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` zapisuje standardowy blok `mcpServers`. Użytkownicy rozszerzenia VS Code: wklej ten sam blok przez Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferowane) lub `config.json` (starsze) | `agentmemory connect continue` tworzy `config.yaml`, jeśli konfiguracji nie ma, lub modyfikuje istniejący `config.json`. Dla istniejącego YAML wypisuje blok do wklejenia pod `mcpServers:`, zachowując komentarze i kotwice. Continue używa tablicy, nie obiektu. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` zapisuje pod `context_servers` (klucz Zed, a NIE `mcpServers`). Zdalne serwery MCP można podłączyć zamiast tego przez `{"url": "..."}`. |
@@ -925,7 +717,6 @@ Każdy kodujący agent AI ma wbudowaną pamięć: Claude Code ma `MEMORY.md`, Cu
|---|---|---|
| Utrzymanie | Ręcznie zarządzane instrukcje | Indeksowana historia sesji i wspomnienia |
| Wyszukiwanie | Wczytuje wszystko do kontekstu | BM25 + wektor + graf (tylko top-K) |
-| Koszt tokenów | 22K+ przy 240 obserwacjach | ~1,900 tokenów (92% mniej) |
| Między agentami | Pliki per agent | MCP + REST (każdy agent) |
| Koordynacja | Brak | Leasy, sygnały, akcje, rutyny |
| Obserwowalność | Ręczne czytanie plików | Podgląd w czasie rzeczywistym na :3113 |
@@ -938,7 +729,7 @@ Każdy kodujący agent AI ma wbudowaną pamięć: Claude Code ma `MEMORY.md`, Cu
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1023,7 +814,7 @@ Wyszukiwanie w trzech strumieniach, łączące trzy sygnały:
Gdy indeks wektorowy jest wypełniony, `mem::search` (stojące za `memory_recall`) używa hybrydowego rankera BM25 + wektor. Bez embeddingów używa BM25. `smart-search` może dodatkowo łączyć strukturalne dopasowania z grafu, gdy dane grafu istnieją, również w trybie bez klucza. Przywołanie lekcji działa na dedykowanym indeksie BM25 w pamięci, zamiast skanować cały korpus przy każdym zapytaniu. Zastąpione wersje pamięci są wykluczone z każdej ścieżki przywołania; łańcuch wersji zachowuje ich historię.
-Wektory przetrwają awarię lub wymuszone zabicie procesu. Indeks wektorowy jest zapisywany w paczkach (buckets) nie częściej niż co `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minut). Każdy wektor dodany lub usunięty w międzyczasie jest też natychmiast zapisywany do małego dziennika oczekujących w magazynie stanu, a przy kolejnym starcie jest odtwarzany bez wywoływania dostawcy embeddingów. Każdy udany zapis opróżnia ten dziennik. Dokumenty, które po odtworzeniu wciąż nie mają wektora, są ponownie embeddowane w tle w paczkach po `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), aż żaden nie zostanie, a zatrzymane uzupełnianie kontynuuje się przy następnym starcie. `/agentmemory/status` i podgląd pokazują rozmiar dziennika oczekujących i stan uzupełniania. Instalacje bez klucza nic nie zapisują.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 od razu tokenizuje grekę, cyrylicę, hebrajski, arabski i akcentowaną łacinę. Dla wspomnień w chińskim / japońskim / koreańskim zainstaluj opcjonalne segmentery (`npm install @node-rs/jieba tiny-segmenter`), aby dzielić ciągi CJK na tokeny na poziomie słów; bez nich agentmemory łagodnie przechodzi na tokenizację całych ciągów i jednorazowo wypisuje wskazówkę na stderr.
@@ -1039,9 +830,9 @@ Zwykła instalacja npm zawiera opcjonalne środowisko uruchomieniowe `@huggingfa
| Dostawca | Model | Koszt | Uwagi |
|---|---|---|---|
-| **Lokalny (zalecany opt-in)** | `all-MiniLM-L6-v2` | Bezpłatny | Lokalnie po pierwszym pobraniu modelu, +8pp recall względem samego BM25 |
+| **Lokalny (zalecany opt-in)** | `all-MiniLM-L6-v2` | Bezpłatny | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Darmowy poziom | 100+ języków, wymiary 768/1536/3072 (MRL), wejście do 2048 tokenów. Zastępuje `text-embedding-004` ([przestarzały, wyłączenie 14 stycznia 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Najwyższa jakość |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Płatny | Zoptymalizowany pod kod |
| Cohere | `embed-english-v3.0` | Darmowy okres próbny | Ogólnego przeznaczenia |
| OpenRouter | Każdy model | Zależnie | Proxy dla wielu modeli |
diff --git a/READMEs/README.pt-BR.md b/READMEs/README.pt-BR.md
index 864862b6..86305330 100644
--- a/READMEs/README.pt-BR.md
+++ b/READMEs/README.pt-BR.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -329,223 +327,17 @@ Notas da última versão: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Precisão de recuperação
-
-**coding-agent-life-v1** (corpus interno, reproduzível em sandbox)
-
-| Adaptador | P@5 | R@5 | Taxa de acerto top-5 | Latência p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Taxa de acerto top-5 de 100% no **teto matemático de P@5** deste corpus (0.240, veja a scorecard). O híbrido recupera todas as sessões gold; o grep perde 1 de 2 golds na query temporal multi-sessão. O ganho é **recall + temporal**, não precisão agregada. Este benchmark é pequeno e esparso em golds; o LongMemEval-S maior abaixo diferencia melhor. Detalhamento completo por tipo + nota de correção: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 perguntas)
+Resultados atuais de recuperação, definições das métricas e etapas de reprodução:
-| Sistema | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Economia de tokens
-
-| Abordagem | Tokens/ano | Custo/ano |
-|---|---|---|
-| Colar contexto completo | 19.5M+ | Impossível (excede a janela) |
-| Resumido por LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embeddings locais | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Modelo de embedding: `all-MiniLM-L6-v2` (local, gratuito, sem API key). Relatórios completos: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Comparativo com concorrentes: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) cobrindo agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproduza localmente:** [`eval/README.md`](../eval/README.md), um harness com adaptadores plugáveis para LongMemEval `_s` (500-Q públicas) e `coding-agent-life-v1` (corpus interno de 15 sessões). Adaptadores grep / vector / agentmemory são pontuados lado a lado, saída em NDJSON, e as scorecards publicadas ficam em [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Combina com [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) e [Graphify](https://github.com/safishamsi/graphify).** Indexação de grafos de código, pipelines de build multiagente e grafos de conhecimento mais amplos sobre docs / PDFs / imagens / vídeos. agentmemory lembra do trabalho; esses três projetos iluminam o resto da camada de contexto. Recipes e tabela de roteamento por pergunta: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Comparativo" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Built-in (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tipo</strong></td>
-<td>Engine de memória + servidor MCP</td>
-<td>API de camada de memória</td>
-<td>Runtime de agente completo</td>
-<td>IA pessoal</td>
-<td>API de memória + app</td>
-<td>Hub de memória de time (proxy de LLM)</td>
-<td>Memória vetorial (OSS)</td>
-<td>Engine de memória (Oracle DB)</td>
-<td>Sistema de memória</td>
-<td>Arquivo estático</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Autorreportado</td>
-<td>PersonaMem 76% (autorreportado)</td>
-<td>~96.6% (autorreportado)</td>
-<td>94.4% (autorreportado)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Captura automática</strong></td>
-<td>12 hooks (esforço manual zero)</td>
-<td>Chamadas manuais a <code>add()</code></td>
-<td>O agente se autoedita</td>
-<td>Manual</td>
-<td>Extração no lado da API</td>
-<td>Interceptação por proxy (troca de base-URL)</td>
-<td>Manual</td>
-<td>Extração via API</td>
-<td>Manual</td>
-<td>Edição manual</td>
-</tr>
-<tr>
-<td><strong>Busca</strong></td>
-<td>BM25 + Vector + Graph (fusão RRF)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semântica</td>
-<td>Vector + RAG</td>
-<td>4 tipos de asset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Somente vector</td>
-<td>Vector + semântica</td>
-<td>Ponderada por decaimento</td>
-<td>Carrega tudo no contexto</td>
-</tr>
-<tr>
-<td><strong>Multiagente</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (sem coordenação)</td>
-<td>Somente dentro do runtime do Letta</td>
-<td>Não</td>
-<td>Não</td>
-<td>Papéis de time + assets compartilhados</td>
-<td>Não</td>
-<td>Somente com escopo</td>
-<td>Compartilhado multiagente</td>
-<td>Arquivos por agente</td>
-</tr>
-<tr>
-<td><strong>Dependência de framework</strong></td>
-<td>Nenhuma (qualquer cliente MCP)</td>
-<td>Nenhuma</td>
-<td>Alta (precisa usar Letta)</td>
-<td>Standalone</td>
-<td>Nenhuma</td>
-<td>Proxy na frente de toda chamada de modelo</td>
-<td>Nenhuma</td>
-<td>Oracle Database</td>
-<td>Nenhuma</td>
-<td>Formato por agente</td>
-</tr>
-<tr>
-<td><strong>Dependências externas</strong></td>
-<td>Nenhum serviço de banco de dados separado necessário</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + BD vetorial</td>
-<td>Várias</td>
-<td>Nuvem gerenciada</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Nenhuma</td>
-<td>Nenhuma</td>
-</tr>
-<tr>
-<td><strong>Ciclo de vida da memória</strong></td>
-<td>Consolidação de 4 níveis + decaimento + auto-esquecimento</td>
-<td>Extração passiva</td>
-<td>Gerenciado pelo agente</td>
-<td>Manual</td>
-<td>Auto-esquecimento</td>
-<td>Revisão manual; roteamento automático em andamento</td>
-<td>Nenhum</td>
-<td>Não informado</td>
-<td>Decaimento + consolidação</td>
-<td>Poda manual</td>
-</tr>
-<tr>
-<td><strong>Eficiência de tokens</strong></td>
-<td>~1,900 tokens/sessão ($10/ano)</td>
-<td>Varia por integração</td>
-<td>Memória principal no contexto</td>
-<td>Varia</td>
-<td>Precificação em nuvem</td>
-<td>Não informado</td>
-<td>Sem orçamento de tokens</td>
-<td>Baseado em LLM (varia)</td>
-<td>Varia</td>
-<td>22K+ tokens em 240 observações</td>
-</tr>
-<tr>
-<td><strong>Viewer em tempo real</strong></td>
-<td>Sim (porta 3113)</td>
-<td>Dashboard na nuvem</td>
-<td>Dashboard na nuvem</td>
-<td>Web UI</td>
-<td>Dashboard na nuvem</td>
-<td>Web UI do Hub</td>
-<td>Não</td>
-<td>Não</td>
-<td>Não</td>
-<td>Não</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Sim (padrão)</td>
-<td>Opcional</td>
-<td>Opcional</td>
-<td>Sim</td>
-<td>Não (somente nuvem)</td>
-<td>Sim (Docker)</td>
-<td>Sim</td>
-<td>Sim (Oracle DB)</td>
-<td>Sim</td>
-<td>Sim</td>
-</tr>
-</table>
-
-<sub>Nota de benchmark: apenas o R@5 do agentmemory é resultado medido por nós mesmos (LongMemEval-S, reproduzível a partir de <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Os números de mem0 e Letta são os números LoCoMo publicados por eles (um dataset diferente); os números de MemPalace, supermemory, TencentDB (PersonaMem) e oracleagentmemory são alegações autorreportadas dos fornecedores que não reproduzimos de forma independente (a execução do oracleagentmemory usou GPT-5.5 contra um Oracle AI Database). Mostrados lado a lado apenas como ordem de grandeza, não como comparação direta sobre dados idênticos. As contagens de stars são aproximadas e mudam com o tempo.</sub>
-
-**Novos entrantes** que vale a pena conhecer, comparados em profundidade em [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistema | ⭐ | Abordagem |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Grafo de conhecimento temporal; os resultados publicados mais fortes em queries temporais (LongMemEval 63.8%), mas o grafo é construído de forma assíncrona, então fatos recentes podem atrasar |
-| Cognee | 30K | Ingestão de documento para grafo de conhecimento, somente Python, feito para extração estruturada de entidades em vez de captura de sessão |
+Comparação atualizada de produtos e opções de implantação, com fontes:
-O AgentMemory combina captura por agentes de programação, histórico de projeto pesquisável e visualizador local. O uso sem chave está disponível desde a primeira execução.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -777,7 +569,7 @@ Guia completo: [`integrations/hermes/`](../integrations/hermes/)
Inicie o servidor de memória: `npx -y @agentmemory/agentmemory@latest`
-#### Skills nativas via `npx skills add` (50+ agentes)
+#### Skills nativas via `npx skills add`
agentmemory entrega 17 skills no formato `<dir>/SKILL.md` estilo Claude Code: 9 skills de ação invocáveis (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) e 8 skills de referência que o agente carrega sob demanda (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). As skills de referência carregam tabelas de dados geradas a partir do código-fonte, então nunca ficam defasadas. O CLI [`skills`](https://npmjs.com/package/skills) da vercel-labs as instala automaticamente no diretório nativo de skills do agente chamador em 50+ agentes (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf e mais):
@@ -834,7 +626,7 @@ A entrada do agentmemory é o **mesmo bloco de servidor MCP** em todo host que u
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` instala o MCP e os hooks de captura no diretório de customização compartilhado. Veja [configuração e limites do Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` usa a mesma configuração de MCP e hooks das versões atuais do IDE. Instalações existentes devem atualizar com `--force`; veja as [notas de atualização](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` escreve a configuração no nível de usuário. Overrides de workspace vão em `.kiro/settings/mcp.json` ao lado do seu código. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escreve o bloco `mcpServers` padrão. O Warp também auto-descobre skills de `.claude/skills/`; instalado o plugin do Claude Code, as 8 skills do agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) aparecem nativamente na paleta de slash commands do Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escreve o bloco `mcpServers` padrão. O Warp também auto-descobre skills de `.claude/skills/`; instalado o plugin do Claude Code, as 9 skills do agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) aparecem nativamente na paleta de slash commands do Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` escreve o bloco `mcpServers` padrão. Usuários da extensão do VS Code: cole o mesmo bloco via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferido) ou `config.json` (legado) | `agentmemory connect continue` cria `config.yaml` se estiver ausente ou modifica o `config.json` existente. Se o YAML já existir, mostra o bloco para adicionar em `mcpServers:` e preservar comentários e âncoras. O Continue usa um array, não um objeto. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` escreve sob `context_servers` (a chave do Zed, NÃO `mcpServers`). Servidores MCP remotos podem ser cabeados via `{"url": "..."}` em vez disso. |
@@ -931,7 +723,6 @@ Todo agente de codificação com IA vem com memória integrada: Claude Code tem
|---|---|---|
| Manutenção | Instruções mantidas manualmente | Histórico de sessões e memórias indexadas |
| Busca | Carrega tudo no contexto | BM25 + vector + graph (somente top-K) |
-| Custo em tokens | 22K+ em 240 observações | ~1,900 tokens (92% menos) |
| Entre agentes | Arquivos por agente | MCP + REST (qualquer agente) |
| Coordenação | Nenhuma | Leases, signals, actions, routines |
| Observabilidade | Leitura manual de arquivos | Viewer em tempo real na :3113 |
@@ -944,7 +735,7 @@ Todo agente de codificação com IA vem com memória integrada: Claude Code tem
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1029,7 +820,7 @@ Combinados com Reciprocal Rank Fusion (RRF, k=60) e diversificados por sessão (
Quando um índice vetorial está populado, o `mem::search` (por trás do `memory_recall`) usa o ranker híbrido BM25 + vector. Sem embeddings ele usa BM25. O `smart-search` pode ainda combinar correspondências estruturais do grafo quando existem dados de grafo, inclusive no modo sem chave. O recall de lessons roda sobre um índice BM25 dedicado em memória, em vez de escanear todo o corpus a cada query. Versões supersedidas de memória são excluídas de todo caminho de recall; a cadeia de versões mantém o histórico delas.
-Vetores sobrevivem a um crash ou force-kill. O índice vetorial é salvo em buckets no máximo a cada `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutos). Todo vetor adicionado ou removido nesse meio tempo também é gravado imediatamente em um pequeno log pendente no state store, e o próximo início o reproduz sem chamar o provider de embedding. Cada save bem-sucedido esvazia o log. Documentos que ainda não têm vetor depois da reprodução são re-embeddados em segundo plano em lotes de `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) até que não reste nenhum, e um backfill interrompido continua no próximo início. `/agentmemory/status` e o viewer mostram o tamanho do log pendente e o estado do backfill. Instalações sem chave não escrevem nada.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
O BM25 tokeniza grego, cirílico, hebraico, árabe e latim acentuado nativamente. Para memórias em chinês / japonês / coreano, instale os segmentadores opcionais (`npm install @node-rs/jieba tiny-segmenter`) para dividir sequências CJK em tokens no nível de palavra; sem eles, o agentmemory cai graciosamente para tokenização da sequência inteira e imprime uma dica única no stderr.
@@ -1045,9 +836,9 @@ A instalação normal via npm inclui o runtime opcional `@huggingface/transforme
| Provider | Modelo | Custo | Notas |
|---|---|---|---|
-| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratuito | No dispositivo após o primeiro download do modelo, +8pp de recall sobre BM25 sozinho |
+| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratuito | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Camada gratuita | 100+ idiomas, dimensões 768/1536/3072 (MRL), entrada de 2048 tokens. Substitui o `text-embedding-004` ([descontinuado, desligamento em 14 de jan. de 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Maior qualidade |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Pago | Otimizado para código |
| Cohere | `embed-english-v3.0` | Trial gratuito | Uso geral |
| OpenRouter | Qualquer modelo | Varia | Proxy multi-modelo |
diff --git a/READMEs/README.pt-PT.md b/READMEs/README.pt-PT.md
index 1c280bcc..94b2335b 100644
--- a/READMEs/README.pt-PT.md
+++ b/READMEs/README.pt-PT.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% de recuperação R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% menos tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 ferramentas MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hooks automáticos" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 bases de dados externas" height="38" /></picture>
@@ -331,223 +329,17 @@ Notas da versão mais recente: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Precisão de Recuperação
-
-**coding-agent-life-v1** (corpus interno, reproduzível em sandbox)
-
-| Adaptador | P@5 | R@5 | Taxa de acerto top-5 | latência p50 |
-|---|---|---|---|---|
-| **agentmemory híbrido** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep de referência | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Taxa de acerto top-5 de 100% no **teto matemático do P@5** para este corpus (0.240, ver scorecard). O híbrido recupera todas as sessões gold; o grep falha 1 de 2 golds na query temporal multi-sessão. O ganho está no **recall + temporal**, não na precisão agregada. Este benchmark é pequeno e escasso em golds; o LongMemEval-S maior, abaixo, diferencia melhor. Detalhamento completo por tipo + nota de correção: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 perguntas)
+Resultados atuais de recuperação, definições das métricas e passos de reprodução:
-| Sistema | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| fallback apenas BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Economia de Tokens
-
-| Abordagem | Tokens/ano | Custo/ano |
-|---|---|---|
-| Colar o contexto completo | 19.5M+ | Impossível (excede a janela) |
-| Resumido por LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embeddings locais | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Modelo de embedding: `all-MiniLM-L6-v2` (local, gratuito, sem chave de API). Relatórios completos: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Comparação com concorrentes: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) cobrindo agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproduzir localmente:** [`eval/README.md`](../eval/README.md), um harness com adaptadores encaixáveis para o LongMemEval `_s` (500 perguntas públicas) + `coding-agent-life-v1` (corpus interno de 15 sessões). Os adaptadores grep / vetor / agentmemory são avaliados lado a lado, com saída NDJSON, e os scorecards publicados ficam em [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Combina-se com [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) e [Graphify](https://github.com/safishamsi/graphify).** Indexação de grafos de código, pipelines de build multiagente e grafos de conhecimento mais amplos em docs / PDFs / imagens / vídeos. O agentmemory recorda o trabalho; estes três projetos iluminam o resto da camada de contexto. Receitas + tabela de encaminhamento de perguntas: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Concorrentes" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Incorporado (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tipo</strong></td>
-<td>Motor de memória + servidor MCP</td>
-<td>API de camada de memória</td>
-<td>Runtime completo de agente</td>
-<td>IA pessoal</td>
-<td>API de memória + app</td>
-<td>Hub de memória de equipa (proxy LLM)</td>
-<td>Memória vetorial (OSS)</td>
-<td>Motor de memória (Oracle DB)</td>
-<td>Sistema de memória</td>
-<td>Ficheiro estático</td>
-</tr>
-<tr>
-<td><strong>Recuperação R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Autorreportado</td>
-<td>PersonaMem 76% (autorreportado)</td>
-<td>~96.6% (autorreportado)</td>
-<td>94.4% (autorreportado)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Captura automática</strong></td>
-<td>12 hooks (esforço manual zero)</td>
-<td>Chamadas manuais a <code>add()</code></td>
-<td>Autoedição pelo agente</td>
-<td>Manual</td>
-<td>Extração do lado da API</td>
-<td>Interceção via proxy (troca de URL base)</td>
-<td>Manual</td>
-<td>Extração via API</td>
-<td>Manual</td>
-<td>Edição manual</td>
-</tr>
-<tr>
-<td><strong>Pesquisa</strong></td>
-<td>BM25 + Vetor + Grafo (fusão RRF)</td>
-<td>Vetor + Grafo</td>
-<td>Vetor (arquivo)</td>
-<td>Semântica</td>
-<td>Vetor + RAG</td>
-<td>4 tipos de recursos (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Apenas vetor</td>
-<td>Vetor + semântica</td>
-<td>Ponderada por decaimento</td>
-<td>Carrega tudo para o contexto</td>
-</tr>
-<tr>
-<td><strong>Multiagente</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (sem coordenação)</td>
-<td>Apenas dentro do runtime da Letta</td>
-<td>Não</td>
-<td>Não</td>
-<td>Papéis de equipa + recursos partilhados</td>
-<td>Não</td>
-<td>Apenas com âmbito definido</td>
-<td>Partilhada multiagente</td>
-<td>Ficheiros por agente</td>
-</tr>
-<tr>
-<td><strong>Dependência de framework</strong></td>
-<td>Nenhuma (qualquer cliente MCP)</td>
-<td>Nenhuma</td>
-<td>Alta (tem de usar Letta)</td>
-<td>Autónomo</td>
-<td>Nenhuma</td>
-<td>O proxy intermedeia todas as chamadas ao modelo</td>
-<td>Nenhuma</td>
-<td>Oracle Database</td>
-<td>Nenhuma</td>
-<td>Formato por agente</td>
-</tr>
-<tr>
-<td><strong>Dependências externas</strong></td>
-<td>Nenhum serviço de base de dados separado necessário</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + BD vetorial</td>
-<td>Várias</td>
-<td>Nuvem gerida</td>
-<td>Stack Docker (Core + Hub + Proxy)</td>
-<td>Armazenamento vetorial</td>
-<td>Oracle AI Database</td>
-<td>Nenhuma</td>
-<td>Nenhuma</td>
-</tr>
-<tr>
-<td><strong>Ciclo de vida da memória</strong></td>
-<td>Consolidação em 4 níveis + decaimento + esquecimento automático</td>
-<td>Extração passiva</td>
-<td>Gerido pelo agente</td>
-<td>Manual</td>
-<td>Esquecimento automático</td>
-<td>Revisão manual; encaminhamento automático em desenvolvimento</td>
-<td>Nenhum</td>
-<td>Não indicado</td>
-<td>Decaimento + consolidação</td>
-<td>Limpeza manual</td>
-</tr>
-<tr>
-<td><strong>Eficiência de tokens</strong></td>
-<td>~1,900 tokens/sessão ($10/ano)</td>
-<td>Varia com a integração</td>
-<td>Memória principal no contexto</td>
-<td>Varia</td>
-<td>Preços na nuvem</td>
-<td>Não indicado</td>
-<td>Sem orçamento de tokens</td>
-<td>Suportado por LLM (varia)</td>
-<td>Varia</td>
-<td>22K+ tokens em 240 observações</td>
-</tr>
-<tr>
-<td><strong>Visualizador em tempo real</strong></td>
-<td>Sim (porta 3113)</td>
-<td>Painel na nuvem</td>
-<td>Painel na nuvem</td>
-<td>Interface Web</td>
-<td>Painel na nuvem</td>
-<td>Interface Web do hub</td>
-<td>Não</td>
-<td>Não</td>
-<td>Não</td>
-<td>Não</td>
-</tr>
-<tr>
-<td><strong>Auto-hospedado</strong></td>
-<td>Sim (predefinição)</td>
-<td>Opcional</td>
-<td>Opcional</td>
-<td>Sim</td>
-<td>Não (apenas na nuvem)</td>
-<td>Sim (Docker)</td>
-<td>Sim</td>
-<td>Sim (Oracle DB)</td>
-<td>Sim</td>
-<td>Sim</td>
-</tr>
-</table>
-
-<sub>Nota sobre o benchmark: apenas o R@5 do agentmemory é um resultado medido por nós (LongMemEval-S, reproduzível a partir de <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Os valores de mem0 e Letta são os números LoCoMo publicados por eles (um dataset diferente); os valores de MemPalace, supermemory, TencentDB (PersonaMem) e oracleagentmemory são afirmações autorreportadas pelos fornecedores que não reproduzimos de forma independente (a execução do oracleagentmemory usou o GPT-5.5 contra uma Oracle AI Database). Mostrados lado a lado apenas como estimativa aproximada, não como comparação direta sobre dados idênticos. As contagens de estrelas são aproximadas e variam com o tempo.</sub>
-
-**Participantes mais recentes** que vale a pena conhecer, comparados em profundidade em [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistema | ⭐ | Abordagem |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Grafo de conhecimento temporal; os resultados publicados mais fortes em queries temporais (LongMemEval 63.8%), mas o grafo é construído de forma assíncrona, por isso os factos recentes podem atrasar |
-| Cognee | 30K | Ingestão de documentos para grafo de conhecimento, apenas Python, construído para extração estruturada de entidades em vez de captura de sessões |
+Comparação atualizada de produtos e opções de implementação, com fontes:
-O AgentMemory combina captura por agentes de programação, histórico de projeto pesquisável e visualizador local. O uso sem chave está disponível desde a primeira execução.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Guia completo: [`integrations/hermes/`](../integrations/hermes/)
Inicie o servidor de memória: `npx -y @agentmemory/agentmemory@latest`
-#### Skills nativas via `npx skills add` (50+ agentes)
+#### Skills nativas via `npx skills add`
O agentmemory distribui 17 skills no formato `<dir>/SKILL.md`, no estilo do Claude Code: 9 skills de ação invocáveis (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) e 8 skills de referência que o agente carrega sob demanda (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). As skills de referência trazem tabelas de dados geradas a partir do código-fonte, por isso nunca ficam desatualizadas. O CLI [`skills`](https://npmjs.com/package/skills) da vercel-labs instala-as automaticamente no diretório nativo de skills do agente que o invoca, em mais de 50 agentes (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, e outros):
@@ -836,7 +628,7 @@ A entrada do agentmemory é o **mesmo bloco de servidor MCP** em todos os anfitr
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` instala o MCP e os hooks de captura no diretório de personalização partilhado. Consulte [configuração e limites do Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` usa a mesma configuração de MCP e hooks das versões atuais do IDE. As instalações existentes devem atualizar com `--force`; consulte as [notas de atualização](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` escreve a configuração ao nível do utilizador. As substituições ao nível do workspace vão em `.kiro/settings/mcp.json`, junto ao seu código. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escreve o bloco `mcpServers` padrão. O Warp também deteta automaticamente skills em `.claude/skills/`; depois de instalado o plugin do Claude Code, as 8 skills do agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) aparecem nativamente na paleta de comandos slash do Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` escreve o bloco `mcpServers` padrão. O Warp também deteta automaticamente skills em `.claude/skills/`; depois de instalado o plugin do Claude Code, as 9 skills do agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) aparecem nativamente na paleta de comandos slash do Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` escreve o bloco `mcpServers` padrão. Utilizadores da extensão VS Code: cole o mesmo bloco através de Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferido) ou `config.json` (legado) | `agentmemory connect continue` cria `config.yaml` se estiver ausente ou modifica o `config.json` existente. Se o YAML já existir, mostra o bloco para adicionar em `mcpServers:` e preservar comentários e âncoras. O Continue usa um array, não um objeto. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` escreve sob `context_servers` (a chave do Zed, NÃO `mcpServers`). Servidores MCP remotos podem, em alternativa, ser ligados através de `{"url": "..."}`. |
@@ -927,7 +719,6 @@ Todo o agente de codificação com IA vem com memória incorporada: o Claude Cod
|---|---|---|
| Manutenção | Instruções mantidas manualmente | Histórico de sessões e memórias indexadas |
| Pesquisa | Carrega tudo para o contexto | BM25 + vetor + grafo (apenas top-K) |
-| Custo em tokens | 22K+ em 240 observações | ~1,900 tokens (92% menos) |
| Entre agentes | Ficheiros por agente | MCP + REST (qualquer agente) |
| Coordenação | Nenhuma | Leases, signals, actions, routines |
| Observabilidade | Ler ficheiros manualmente | Visualizador em tempo real na :3113 |
@@ -940,7 +731,7 @@ Todo o agente de codificação com IA vem com memória incorporada: o Claude Cod
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ Combinados com Reciprocal Rank Fusion (RRF, k=60) e diversificados por sessão (
Quando um índice vetorial está preenchido, o `mem::search` (por trás de `memory_recall`) usa o ranker híbrido BM25 + vetor. Sem embeddings, usa o BM25. O `smart-search` pode ainda combinar correspondências estruturais do grafo quando existem dados de grafo, inclusive em modo sem chave. O recall de lessons corre sobre um índice BM25 dedicado em memória, em vez de percorrer todo o corpus a cada query. As versões de memória substituídas são excluídas de todos os caminhos de recall; a cadeia de versões mantém o seu histórico.
-Os vetores sobrevivem a um crash ou a uma paragem forçada. O índice vetorial é guardado em blocos no máximo a cada `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minutos). Cada vetor adicionado ou removido entretanto também é escrito imediatamente num pequeno registo pendente no state store, e o arranque seguinte reproduz esse registo sem chamar o fornecedor de embeddings. Cada gravação bem-sucedida esvazia o registo. Os documentos que ainda não têm vetor depois da reprodução são re-embutidos (re-embedded) em segundo plano, em lotes de `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), até não restar nenhum, e um preenchimento retroativo (backfill) interrompido continua no arranque seguinte. O `/agentmemory/status` e o visualizador mostram o tamanho do registo pendente e o estado do backfill. As instalações sem chave não escrevem nada.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
O BM25 tokeniza grego, cirílico, hebraico, árabe e latim acentuado de fábrica. Para memórias em chinês / japonês / coreano, instale os segmentadores opcionais (`npm install @node-rs/jieba tiny-segmenter`) para dividir sequências CJK em tokens ao nível da palavra; sem eles, o agentmemory recua suavemente para a tokenização da sequência completa e imprime uma dica única no stderr.
@@ -1041,9 +832,9 @@ A instalação normal via npm inclui o runtime opcional `@huggingface/transforme
| Fornecedor | Modelo | Custo | Notas |
|---|---|---|---|
-| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratuito | No dispositivo após o primeiro download do modelo, +8pp de recall sobre apenas BM25 |
+| **Local (opt-in recomendado)** | `all-MiniLM-L6-v2` | Gratuito | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Nível gratuito | 100+ idiomas, dimensões 768/1536/3072 (MRL), entrada de 2048 tokens. Substitui o `text-embedding-004` ([descontinuado, encerramento a 14 de janeiro de 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Qualidade mais alta |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Pago | Otimizado para código |
| Cohere | `embed-english-v3.0` | Teste gratuito | Uso geral |
| OpenRouter | Qualquer modelo | Varia | Proxy multimodelo |
diff --git a/READMEs/README.ro-RO.md b/READMEs/README.ro-RO.md
index b2401fe6..1cefbf86 100644
--- a/READMEs/README.ro-RO.md
+++ b/READMEs/README.ro-RO.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% regăsire R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% mai puțini tokeni" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 de instrumente MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hook-uri automate" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 baze de date externe" height="38" /></picture>
@@ -333,223 +331,17 @@ Notele celei mai recente versiuni: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmark-uri" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Acuratețea regăsirii
-
-**coding-agent-life-v1** (corpus intern, reproductibil în sandbox)
-
-| Adaptor | P@5 | R@5 | Rata de succes Top-5 | Latență p50 |
-|---|---|---|---|---|
-| **agentmemory hibrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep (referință) | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Rată de succes Top-5 de 100% la **plafonul matematic P@5** pentru acest corpus (0.240, vezi scorecard-ul). Modul hibrid regăsește fiecare sesiune de referință (gold); grep ratează 1 din 2 sesiuni de referință la interogarea temporală pe mai multe sesiuni. Avantajul este pe **recall + temporal**, nu pe precizia agregată. Acest benchmark este mic și are puține exemple de referință; LongMemEval-S, mai amplu, de mai jos diferențiază mai bine. Detalierea completă pe tipuri + nota de corecție: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 de întrebări)
+Rezultatele actuale ale căutării, definițiile metricilor și pașii de reproducere:
-| Sistem | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Doar BM25 (fallback) | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Economii de tokeni
-
-| Abordare | Tokeni/an | Cost/an |
-|---|---|---|
-| Lipire context complet | 19.5M+ | Imposibil (depășește fereastra de context) |
-| Rezumat de LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embeddings locale | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Model de embedding: `all-MiniLM-L6-v2` (local, gratuit, fără cheie API). Rapoarte complete: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Comparație cu competitorii: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md), care acoperă agentmemory vs mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproduce local:** [`eval/README.md`](../eval/README.md), un harness cu adaptoare interschimbabile pentru LongMemEval `_s` (500 de întrebări publice) + `coding-agent-life-v1` (corpus intern de 15 sesiuni). Adaptoarele grep / vector / agentmemory sunt scorate în paralel, cu output NDJSON; scorecard-urile publicate ajung în [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Se combină bine cu [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) și [Graphify](https://github.com/safishamsi/graphify).** Indexare code-graph, pipeline-uri de build multi-agent și grafuri de cunoștințe mai ample pe documente / PDF-uri / imagini / videoclipuri. agentmemory reține munca; aceste trei proiecte iluminează restul stratului de context. Rețete + tabel de direcționare a întrebărilor: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitori" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Încorporat (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tip</strong></td>
-<td>Motor de memorie + server MCP</td>
-<td>API strat de memorie</td>
-<td>Runtime complet de agent</td>
-<td>AI personal</td>
-<td>API de memorie + aplicație</td>
-<td>Hub de memorie de echipă (proxy LLM)</td>
-<td>Memorie vectorială (OSS)</td>
-<td>Motor de memorie (Oracle DB)</td>
-<td>Sistem de memorie</td>
-<td>Fișier static</td>
-</tr>
-<tr>
-<td><strong>Regăsire R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Auto-raportat</td>
-<td>PersonaMem 76% (auto-raportat)</td>
-<td>~96.6% (auto-raportat)</td>
-<td>94.4% (auto-raportat)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Captare automată</strong></td>
-<td>12 hook-uri (fără efort manual)</td>
-<td>Apeluri manuale <code>add()</code></td>
-<td>Auto-editare de către agent</td>
-<td>Manual</td>
-<td>Extracție pe partea de API</td>
-<td>Interceptare prin proxy (schimbare base-URL)</td>
-<td>Manual</td>
-<td>Extracție API</td>
-<td>Manual</td>
-<td>Editare manuală</td>
-</tr>
-<tr>
-<td><strong>Căutare</strong></td>
-<td>BM25 + Vector + Graf (fuziune RRF)</td>
-<td>Vector + Graf</td>
-<td>Vector (arhivă)</td>
-<td>Semantică</td>
-<td>Vector + RAG</td>
-<td>4 tipuri de active (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Doar vector</td>
-<td>Vector + semantică</td>
-<td>Ponderată prin declin (decay)</td>
-<td>Încarcă totul în context</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + lease-uri + semnale</td>
-<td>API (fără coordonare)</td>
-<td>Doar în interiorul runtime-ului Letta</td>
-<td>Nu</td>
-<td>Nu</td>
-<td>Roluri de echipă + active partajate</td>
-<td>Nu</td>
-<td>Doar cu scope limitat</td>
-<td>Partajat multi-agent</td>
-<td>Fișiere per agent</td>
-</tr>
-<tr>
-<td><strong>Dependență de framework</strong></td>
-<td>Niciuna (orice client MCP)</td>
-<td>Niciuna</td>
-<td>Ridicată (trebuie folosit Letta)</td>
-<td>Independent (standalone)</td>
-<td>Niciuna</td>
-<td>Proxy-ul intermediază fiecare apel de model</td>
-<td>Niciuna</td>
-<td>Oracle Database</td>
-<td>Niciuna</td>
-<td>Format per agent</td>
-</tr>
-<tr>
-<td><strong>Dependențe externe</strong></td>
-<td>Nu necesită un serviciu de baze de date separat</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + bază de date vectorială</td>
-<td>Multiple</td>
-<td>Cloud administrat</td>
-<td>Stivă Docker (Core + Hub + Proxy)</td>
-<td>Magazin vectorial</td>
-<td>Oracle AI Database</td>
-<td>Niciuna</td>
-<td>Niciuna</td>
-</tr>
-<tr>
-<td><strong>Ciclul de viață al memoriei</strong></td>
-<td>Consolidare pe 4 niveluri + declin + uitare automată</td>
-<td>Extracție pasivă</td>
-<td>Gestionat de agent</td>
-<td>Manual</td>
-<td>Uitare automată</td>
-<td>Revizuire manuală; direcționare automată în lucru</td>
-<td>Niciuna</td>
-<td>Nespecificat</td>
-<td>Declin + consolidare</td>
-<td>Curățare manuală</td>
-</tr>
-<tr>
-<td><strong>Eficiența tokenilor</strong></td>
-<td>~1,900 tokeni/sesiune ($10/an)</td>
-<td>Variază în funcție de integrare</td>
-<td>Memorie de bază (core) în context</td>
-<td>Variază</td>
-<td>Preț cloud</td>
-<td>Nespecificat</td>
-<td>Fără buget de tokeni</td>
-<td>Susținut de LLM (variază)</td>
-<td>Variază</td>
-<td>22K+ tokeni la 240 observații</td>
-</tr>
-<tr>
-<td><strong>Vizualizator în timp real</strong></td>
-<td>Da (port 3113)</td>
-<td>Dashboard cloud</td>
-<td>Dashboard cloud</td>
-<td>Interfață web</td>
-<td>Dashboard cloud</td>
-<td>Interfață web Hub</td>
-<td>Nu</td>
-<td>Nu</td>
-<td>Nu</td>
-<td>Nu</td>
-</tr>
-<tr>
-<td><strong>Auto-hostat</strong></td>
-<td>Da (implicit)</td>
-<td>Opțional</td>
-<td>Opțional</td>
-<td>Da</td>
-<td>Nu (doar cloud)</td>
-<td>Da (Docker)</td>
-<td>Da</td>
-<td>Da (Oracle DB)</td>
-<td>Da</td>
-<td>Da</td>
-</tr>
-</table>
-
-<sub>Notă despre benchmark: doar R@5 pentru agentmemory este un rezultat măsurat de noi înșine (LongMemEval-S, reproductibil din <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Cifrele pentru mem0 și Letta sunt numerele lor publicate pe LoCoMo (un set de date diferit); cifrele pentru MemPalace, supermemory, TencentDB (PersonaMem) și oracleagentmemory sunt afirmații auto-raportate de furnizori, pe care nu le-am reprodus independent (rularea oracleagentmemory a folosit GPT-5.5 împotriva unei Oracle AI Database). Prezentate una lângă alta doar orientativ, nu ca o comparație directă pe date identice. Numărul de stele este aproximativ și variază în timp.</sub>
-
-**Jucători mai noi** merită cunoscuți, comparați în detaliu în [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistem | ⭐ | Unghi |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Graf de cunoștințe temporal; cele mai bune rezultate publicate pentru interogări temporale (LongMemEval 63.8%), dar graful se construiește asincron, așa că faptele noi pot întârzia |
-| Cognee | 30K | Ingestie document-în-graf-de-cunoștințe, doar Python, construit pentru extracție structurată de entități, nu pentru captarea sesiunilor |
+Comparație actualizată a produselor și opțiunilor de implementare, cu surse:
-AgentMemory combină captura sesiunilor, istoricul căutabil al proiectului și un viewer local, cu funcționare fără cheie de la prima pornire.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Ghid complet: [`integrations/hermes/`](../integrations/hermes/)
Pornește serverul de memorie: `npx -y @agentmemory/agentmemory@latest`
-#### Skill-uri native prin `npx skills add` (50+ agenți)
+#### Skill-uri native prin `npx skills add`
agentmemory vine cu 17 skill-uri în formatul `<dir>/SKILL.md`, în stilul Claude Code: 9 skill-uri de acțiune invocabile (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) și 8 skill-uri de referință pe care agentul le încarcă la cerere (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Skill-urile de referință conțin tabele de date generate din sursă, așa că nu devin niciodată desincronizate. CLI-ul [`skills`](https://npmjs.com/package/skills) de la vercel-labs le instalează automat în directorul nativ de skill-uri al agentului apelant, pe 50+ agenți (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf și altele):
@@ -844,7 +636,7 @@ Intrarea agentmemory este **același bloc de server MCP** pe fiecare host care f
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` instalează MCP și hook-urile de captare în directorul partajat de personalizare. Vezi [configurarea și limitele Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` folosește aceeași configurație MCP și de hook-uri ca versiunile IDE actuale. Instalările existente ar trebui să se reîmprospăteze cu `--force`; vezi [notele de actualizare](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` scrie configurația la nivel de utilizator. Suprascrierile la nivel de workspace se pun în `.kiro/settings/mcp.json`, alături de codul tău. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` scrie blocul standard `mcpServers`. Warp detectează automat și skill-urile din `.claude/skills/`; odată instalat plugin-ul Claude Code, cele 8 skill-uri agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) apar nativ în paleta de comenzi slash a Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` scrie blocul standard `mcpServers`. Warp detectează automat și skill-urile din `.claude/skills/`; odată instalat plugin-ul Claude Code, cele 9 skill-uri agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) apar nativ în paleta de comenzi slash a Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` scrie blocul standard `mcpServers`. Utilizatorii extensiei VS Code: lipește același bloc prin Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferat) sau `config.json` (vechi/legacy) | `agentmemory connect continue` creează `config.yaml` când lipsește configurarea sau modifică `config.json` existent. Pentru YAML existent afișează blocul de adăugat la `mcpServers:`, păstrând comentariile și ancorele. Continue folosește o listă, nu un obiect. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` scrie sub `context_servers` (cheia specifică Zed, NU `mcpServers`). Serverele MCP la distanță pot fi conectate, în schimb, prin `{"url": "..."}`. |
@@ -921,7 +713,6 @@ Fiecare agent de programare AI vine cu memorie încorporată: Claude Code are `M
|---|---|---|
| Întreținere | Instrucțiuni administrate manual | Istoric de sesiuni și memorie indexate |
| Căutare | Încarcă totul în context | BM25 + vector + graf (doar top-K) |
-| Cost în tokeni | 22K+ la 240 de observații | ~1,900 tokeni (92% mai puțin) |
| Între agenți | Fișiere per agent | MCP + REST (orice agent) |
| Coordonare | Niciuna | Lease-uri, semnale, acțiuni, rutine |
| Observabilitate | Citire manuală a fișierelor | Vizualizator în timp real pe :3113 |
@@ -934,7 +725,7 @@ Fiecare agent de programare AI vine cu memorie încorporată: Claude Code are `M
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1019,7 +810,7 @@ Fuzionate prin Reciprocal Rank Fusion (RRF, k=60) și diversificate pe sesiune (
Când indexul vectorial este populat, `mem::search` (din spatele lui `memory_recall`) folosește clasificatorul hibrid BM25 + vector. Fără embeddings, folosește BM25. `smart-search` poate fuziona, în plus, potriviri structurale din graf, atunci când există date de graf, inclusiv în modul keyless. Reamintirea lecțiilor rulează pe un index BM25 dedicat, în memorie, în loc să scaneze întregul corpus la fiecare interogare. Versiunile de memorie suprascrise sunt excluse din fiecare rută de reamintire; lanțul de versiuni le păstrează istoricul.
-Vectorii supraviețuiesc unui crash sau unei terminări forțate (force-kill). Indexul vectorial este salvat pe loturi (buckets) cel mult o dată la fiecare `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minute). Fiecare vector adăugat sau eliminat între timp este scris imediat și într-un mic jurnal de așteptare (pending log) din state store, iar următoarea pornire îl reaplică fără a apela furnizorul de embedding. Fiecare salvare reușită golește jurnalul. Documentele care încă nu au un vector după reaplicare sunt reîncorporate (re-embedded) în fundal, pe loturi de `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), până când nu mai rămâne niciunul, iar o reumplere (backfill) întreruptă continuă la următoarea pornire. `/agentmemory/status` și vizualizatorul arată dimensiunea jurnalului de așteptare și starea reumplerii. Instalările keyless nu scriu nimic.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenizează, din oficiu, greaca, chirilica, ebraica, araba și latina cu diacritice. Pentru memorii în chineză / japoneză / coreeană, instalează segmentatoarele opționale (`npm install @node-rs/jieba tiny-segmenter`) pentru a împărți secvențele CJK în tokeni la nivel de cuvânt; fără ele, agentmemory trece ușor la tokenizarea întregii secvențe și afișează o singură dată un indiciu pe stderr.
@@ -1035,9 +826,9 @@ Instalarea npm normală include runtime-ul opțional `@huggingface/transformers`
| Furnizor | Model | Cost | Note |
|---|---|---|---|
-| **Local (opțiune recomandată)** | `all-MiniLM-L6-v2` | Gratuit | Pe dispozitiv, după prima descărcare a modelului; +8pp recall față de doar-BM25 |
+| **Local (opțiune recomandată)** | `all-MiniLM-L6-v2` | Gratuit | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Nivel gratuit | 100+ limbi, 768/1536/3072 dimensiuni (MRL), input de 2048 de tokeni. Înlocuiește `text-embedding-004` ([depreciat, dezactivat 14 ian. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Cea mai bună calitate |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Plătit | Optimizat pentru cod |
| Cohere | `embed-english-v3.0` | Perioadă de probă gratuită | Scop general |
| OpenRouter | Orice model | Variabil | Proxy multi-model |
diff --git a/READMEs/README.ru-RU.md b/READMEs/README.ru-RU.md
index 83bbaf21..33558b58 100644
--- a/READMEs/README.ru-RU.md
+++ b/READMEs/README.ru-RU.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -333,223 +331,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Бенчмарки" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Точность извлечения
-
-**coding-agent-life-v1** (собственный корпус, воспроизводимый в sandbox)
-
-| Адаптер | P@5 | R@5 | Top-5 hit rate | p50 задержка |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% попаданий в top-5 на **математическом потолке P@5** для этого корпуса (0.240, см. scorecard). Hybrid извлекает каждую gold-сессию; grep промахивается на 1 из 2 gold в мультисессионном темпоральном запросе. Прирост даёт **recall + temporal**, а не суммарная точность. Этот бенчмарк маленький и разреженный по gold; более крупный LongMemEval-S ниже различает лучше. Полная разбивка по типам и заметка о поправке: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 вопросов)
+Актуальные результаты поиска, определения метрик и шаги воспроизведения:
-| Система | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Только BM25 (fallback) | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Экономия токенов
-
-| Подход | Токенов/год | Стоимость/год |
-|---|---|---|
-| Вставка полного контекста | 19.5M+ | Невозможно (превышает окно) |
-| Сжатие LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + локальные эмбеддинги | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Модель эмбеддингов: `all-MiniLM-L6-v2` (локальная, бесплатная, без API-ключа). Полные отчёты: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Сравнение с конкурентами: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) — сравнивает agentmemory с mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Воспроизвести локально:** [`eval/README.md`](../eval/README.md) — харнесс со сменными адаптерами для LongMemEval `_s` (публичные 500 вопросов) + `coding-agent-life-v1` (собственный корпус из 15 сессий). Адаптеры grep / вектор / agentmemory оцениваются бок о бок, вывод в NDJSON, опубликованные scorecard-отчёты попадают в [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Сочетается с [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) и [Graphify](https://github.com/safishamsi/graphify).** Индексация графа кода, мультиагентные пайплайны сборки и более широкие графы знаний по докам / PDF / изображениям / видео. agentmemory запоминает работу; эти три проекта оживляют остальную часть контекстного слоя. Рецепты и таблица маршрутизации вопросов: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Сравнение с конкурентами" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Встроенное (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Тип</strong></td>
-<td>Движок памяти + MCP-сервер</td>
-<td>API уровня памяти</td>
-<td>Полноценный агентский runtime</td>
-<td>Персональный ИИ</td>
-<td>API памяти + приложение</td>
-<td>Командный хаб памяти (LLM-прокси)</td>
-<td>Векторная память (OSS)</td>
-<td>Движок памяти (Oracle DB)</td>
-<td>Система памяти</td>
-<td>Статический файл</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Заявлено вендором</td>
-<td>PersonaMem 76% (заявлено вендором)</td>
-<td>~96.6% (заявлено вендором)</td>
-<td>94.4% (заявлено вендором)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Авто-захват</strong></td>
-<td>12 хуков (никаких ручных усилий)</td>
-<td>Ручные вызовы <code>add()</code></td>
-<td>Агент сам редактирует</td>
-<td>Вручную</td>
-<td>Извлечение на стороне API</td>
-<td>Перехват через прокси (подмена base-URL)</td>
-<td>Вручную</td>
-<td>Извлечение через API</td>
-<td>Вручную</td>
-<td>Ручное редактирование</td>
-</tr>
-<tr>
-<td><strong>Поиск</strong></td>
-<td>BM25 + Vector + Graph (RRF-слияние)</td>
-<td>Vector + Graph</td>
-<td>Vector (архивный)</td>
-<td>Семантический</td>
-<td>Vector + RAG</td>
-<td>4 типа ассетов (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Только Vector</td>
-<td>Vector + семантический</td>
-<td>Взвешенный по затуханию</td>
-<td>Загружает всё в контекст</td>
-</tr>
-<tr>
-<td><strong>Мультиагентность</strong></td>
-<td>MCP + REST + lease'ы + сигналы</td>
-<td>API (без координации)</td>
-<td>Только внутри runtime Letta</td>
-<td>Нет</td>
-<td>Нет</td>
-<td>Командные роли + общие ассеты</td>
-<td>Нет</td>
-<td>Только scope'ы</td>
-<td>Мультиагентная общая</td>
-<td>Отдельные файлы на агента</td>
-</tr>
-<tr>
-<td><strong>Привязка к фреймворку</strong></td>
-<td>Нет (любой MCP-клиент)</td>
-<td>Нет</td>
-<td>Высокая (нужен Letta)</td>
-<td>Standalone</td>
-<td>Нет</td>
-<td>Прокси стоит перед каждым вызовом модели</td>
-<td>Нет</td>
-<td>Oracle Database</td>
-<td>Нет</td>
-<td>Формат на агента</td>
-</tr>
-<tr>
-<td><strong>Внешние зависимости</strong></td>
-<td>Отдельный сервис базы данных не требуется</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + векторная БД</td>
-<td>Несколько</td>
-<td>Managed-облако</td>
-<td>Docker-стек (Core + Hub + Proxy)</td>
-<td>Векторное хранилище</td>
-<td>Oracle AI Database</td>
-<td>Нет</td>
-<td>Нет</td>
-</tr>
-<tr>
-<td><strong>Жизненный цикл памяти</strong></td>
-<td>4-уровневая консолидация + затухание + авто-забывание</td>
-<td>Пассивное извлечение</td>
-<td>Управляется агентом</td>
-<td>Вручную</td>
-<td>Авто-забывание</td>
-<td>Ручной ревью; авто-маршрутизация в разработке</td>
-<td>Нет</td>
-<td>Не указано</td>
-<td>Затухание + консолидация</td>
-<td>Ручное усечение</td>
-</tr>
-<tr>
-<td><strong>Эффективность по токенам</strong></td>
-<td>~1,900 токенов/сессия ($10/год)</td>
-<td>Зависит от интеграции</td>
-<td>Core memory в контексте</td>
-<td>Зависит</td>
-<td>Облачные тарифы</td>
-<td>Не указано</td>
-<td>Без токен-бюджета</td>
-<td>На основе LLM (варьируется)</td>
-<td>Зависит</td>
-<td>22K+ токенов при 240 наблюдениях</td>
-</tr>
-<tr>
-<td><strong>Просмотрщик в реальном времени</strong></td>
-<td>Да (порт 3113)</td>
-<td>Облачная панель</td>
-<td>Облачная панель</td>
-<td>Веб-UI</td>
-<td>Облачная панель</td>
-<td>Веб-UI хаба</td>
-<td>Нет</td>
-<td>Нет</td>
-<td>Нет</td>
-<td>Нет</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Да (по умолчанию)</td>
-<td>Опционально</td>
-<td>Опционально</td>
-<td>Да</td>
-<td>Нет (только облако)</td>
-<td>Да (Docker)</td>
-<td>Да</td>
-<td>Да (Oracle DB)</td>
-<td>Да</td>
-<td>Да</td>
-</tr>
-</table>
-
-<sub>Заметка о бенчмарках: только R@5 agentmemory — наш собственный замер (LongMemEval-S, воспроизводимо из <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Цифры mem0 и Letta — их опубликованные результаты LoCoMo (другой датасет); цифры MemPalace, supermemory, TencentDB (PersonaMem) и oracleagentmemory — самозаявленные вендорами значения, которые мы независимо не воспроизводили (прогон oracleagentmemory использовал GPT-5.5 против Oracle AI Database). Показаны рядом только для ориентира, это не сравнение лоб в лоб на одинаковых данных. Количество звёзд приблизительно и меняется со временем.</sub>
-
-**Более новые участники**, которых стоит знать, подробно сравниваются в [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Система | ⭐ | Особенность |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Темпоральный граф знаний; сильнейшие опубликованные результаты на темпоральных запросах (LongMemEval 63.8%), но граф строится асинхронно, поэтому свежие факты могут запаздывать |
-| Cognee | 30K | Превращение документов в граф знаний, только Python, создан для структурированного извлечения сущностей, а не для захвата сессий |
+Актуальное сравнение продуктов и вариантов развёртывания с источниками:
-AgentMemory объединяет захват сессий, поиск по истории проекта и локальный viewer; работа без ключа доступна с первого запуска.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
Запустите сервер памяти: `npx -y @agentmemory/agentmemory@latest`
-#### Нативные skill'ы через `npx skills add` (50+ агентов)
+#### Нативные skill'ы через `npx skills add`
agentmemory поставляет 17 skill'ов в формате `<dir>/SKILL.md` в стиле Claude Code: 9 вызываемых action-skill'ов (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) и 8 справочных skill'ов, которые агент подгружает по мере надобности (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Справочные skill'ы содержат таблицы данных, сгенерированные из исходников, поэтому они никогда не устаревают. CLI [`skills`](https://npmjs.com/package/skills) от vercel-labs автоматически устанавливает их в нативный каталог skill'ов вызывающего агента для 50+ агентов (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf и другие):
@@ -844,7 +636,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` устанавливает MCP и хуки захвата в общем каталоге настроек. См. [установку и ограничения Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` использует ту же конфигурацию MCP и хуков, что и текущие версии IDE. Существующие установки следует обновить с `--force`; см. [заметки об обновлении](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` записывает конфиг на уровне пользователя. Переопределения на уровне workspace — в `.kiro/settings/mcp.json` рядом с кодом. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` записывает стандартный блок `mcpServers`. Warp также автоматически обнаруживает skill'ы из `.claude/skills/`; как только установлен плагин Claude Code, 8 skill'ов agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) нативно появляются в палитре slash-команд Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` записывает стандартный блок `mcpServers`. Warp также автоматически обнаруживает skill'ы из `.claude/skills/`; как только установлен плагин Claude Code, 9 skill'ов agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) нативно появляются в палитре slash-команд Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` записывает стандартный блок `mcpServers`. Пользователи расширения VS Code: вставьте тот же блок через Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (предпочтительно) или `config.json` (legacy) | `agentmemory connect continue` создаёт `config.yaml`, если настройки отсутствуют, или меняет существующий `config.json`. Для существующего YAML выводит блок для `mcpServers:`, сохраняя комментарии и якоря. Continue использует массив, а не объект. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` пишет под `context_servers` (ключ Zed, НЕ `mcpServers`). Удалённые MCP-серверы можно вместо этого подключить через `{"url": "..."}`. |
@@ -927,7 +719,6 @@ REST использует `3111`, viewer `3113`. Оставьте их в при
|---|---|---|
| Обслуживание | Инструкции, редактируемые вручную | Индексируемая история сессий и воспоминания |
| Поиск | Загружает всё в контекст | BM25 + вектор + граф (только top-K) |
-| Цена в токенах | 22K+ при 240 наблюдениях | ~1,900 токенов (на 92% меньше) |
| Между агентами | Файлы на каждого агента | MCP + REST (любой агент) |
| Координация | Нет | Lease'ы, сигналы, action'ы, routine'ы |
| Наблюдаемость | Читать файлы вручную | Просмотрщик в реальном времени на :3113 |
@@ -940,7 +731,7 @@ REST использует `3111`, viewer `3113`. Оставьте их в при
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ SessionStart hook fires
Когда векторный индекс заполнен, `mem::search` (за которым стоит `memory_recall`) использует гибридный ранжировщик BM25 + вектор. Без эмбеддингов используется BM25. `smart-search` может дополнительно сливать структурные совпадения из графа, если граф-данные существуют, в том числе в режиме без ключей. Recall уроков работает на выделенном in-memory BM25-индексе вместо сканирования всего корпуса на каждый запрос. Вытесненные (superseded) версии памяти исключаются из каждого пути recall; цепочка версий хранит их историю.
-Векторы переживают крэш или force-kill. Векторный индекс сохраняется бакетами не чаще, чем раз в `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 минут). Каждый добавленный или удалённый в промежутке вектор также сразу записывается в небольшой отложенный лог в state store, и при следующем запуске он воспроизводится без обращения к провайдеру эмбеддингов. Каждое успешное сохранение очищает лог. Документы, у которых после воспроизведения всё ещё нет вектора, переэмбеддятся в фоне партиями по `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), пока не останется ни одного, а остановленный backfill продолжается при следующем запуске. `/agentmemory/status` и просмотрщик показывают размер отложенного лога и состояние backfill'а. Установки без ключей ничего не пишут.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 «из коробки» токенизирует греческий, кириллицу, иврит, арабский и латиницу с диакритикой. Для записей на китайском / японском / корейском поставьте опциональные сегментаторы (`npm install @node-rs/jieba tiny-segmenter`), чтобы разбивать CJK-последовательности на токены уровня слова; без них agentmemory мягко откатывается к токенизации целых последовательностей и один раз выводит подсказку в stderr.
@@ -1041,9 +832,9 @@ EMBEDDING_PROVIDER=local
| Провайдер | Модель | Стоимость | Заметки |
|---|---|---|---|
-| **Локально (рекомендуется для подключения)** | `all-MiniLM-L6-v2` | Бесплатно | На устройстве после первой загрузки модели, +8 пп recall по сравнению только с BM25 |
+| **Локально (рекомендуется для подключения)** | `all-MiniLM-L6-v2` | Бесплатно | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Бесплатный тариф | 100+ языков, размерности 768/1536/3072 (MRL), вход 2048 токенов. Заменяет `text-embedding-004` ([устарел, отключение 14 янв. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Высочайшее качество |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Платно | Оптимизирован под код |
| Cohere | `embed-english-v3.0` | Бесплатная пробная версия | Общего назначения |
| OpenRouter | Любая модель | Зависит | Мульти-модельный прокси |
diff --git a/READMEs/README.sv-SE.md b/READMEs/README.sv-SE.md
index 3aa89988..760ca650 100644
--- a/READMEs/README.sv-SE.md
+++ b/READMEs/README.sv-SE.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% hämtning R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% färre tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP-verktyg" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 automatiska hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 externa databaser" height="38" /></picture>
@@ -331,223 +329,17 @@ Senaste versionsanteckningarna: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Hämtningsnoggrannhet
-
-**coding-agent-life-v1** (internt korpus, reproducerbart i sandbox)
-
-| Adapter | P@5 | R@5 | Topp-5-träffrate | p50-latens |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep-baslinje | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% topp-5-träffrate vid **P@5-mattaket** för detta korpus (0.240, se scorecard). Hybrid hämtar varje gold-session; grep missar 1 av 2 gold på den flersessions-temporala frågan. Lyftet ligger i **recall + temporal**, inte i aggregerad precision. Detta benchmark är litet och gold-glest; det större LongMemEval-S nedan differentierar bättre. Fullständig uppdelning per typ + rättelsenotis: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 frågor)
+Aktuella sökresultat, definitioner av mått och steg för reproduktion:
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Endast BM25-fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tokenbesparingar
-
-| Metod | Tokens/år | Kostnad/år |
-|---|---|---|
-| Klistra in hela kontexten | 19.5M+ | Omöjligt (överskrider fönstret) |
-| LLM-sammanfattat | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + lokala embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding-modell: `all-MiniLM-L6-v2` (lokal, kostnadsfri, inget API-nyckel krävs). Fullständiga rapporter: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Konkurrentjämförelse: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) som täcker agentmemory mot mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Reproducera lokalt:** [`eval/README.md`](../eval/README.md), en adapter-pluggbar testrigg för LongMemEval `_s` (publikt 500-frågors) + `coding-agent-life-v1` (internt 15-sessionskorpus). Grep-, vektor- och agentmemory-adaptrar poängsätts sida vid sida, NDJSON-utdata, publicerade scorecards hamnar i [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Fungerar bra ihop med [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) och [Graphify](https://github.com/safishamsi/graphify).** Kodgraf-indexering, multiagent-byggpipelines och bredare kunskapsgrafer över dokument/PDF-filer/bilder/videor. agentmemory minns arbetet; dessa tre projekt lyser upp resten av kontextlagret. Recept + frågerouting-tabell: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Jämfört med konkurrenter" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Inbyggt (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Typ</strong></td>
-<td>Minnesmotor + MCP-server</td>
-<td>Minneslager-API</td>
-<td>Fullständig agent-körtid</td>
-<td>Personlig AI</td>
-<td>Minnes-API + app</td>
-<td>Teamminnesnav (LLM-proxy)</td>
-<td>Vektorminne (OSS)</td>
-<td>Minnesmotor (Oracle DB)</td>
-<td>Minnessystem</td>
-<td>Statisk fil</td>
-</tr>
-<tr>
-<td><strong>Hämtning R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Självrapporterat</td>
-<td>PersonaMem 76% (självrapporterat)</td>
-<td>~96.6% (självrapporterat)</td>
-<td>94.4% (självrapporterat)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Automatisk infångning</strong></td>
-<td>12 hooks (noll manuellt arbete)</td>
-<td>Manuella <code>add()</code>-anrop</td>
-<td>Agenten redigerar sig själv</td>
-<td>Manuellt</td>
-<td>Extraktion på API-sidan</td>
-<td>Proxy-avlyssning (byte av bas-URL)</td>
-<td>Manuellt</td>
-<td>API-extraktion</td>
-<td>Manuellt</td>
-<td>Manuell redigering</td>
-</tr>
-<tr>
-<td><strong>Sökning</strong></td>
-<td>BM25 + vektor + graf (RRF-fusion)</td>
-<td>Vektor + graf</td>
-<td>Vektor (arkiverat)</td>
-<td>Semantisk</td>
-<td>Vektor + RAG</td>
-<td>4 typer av tillgångar (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Endast vektor</td>
-<td>Vektor + semantisk</td>
-<td>Avklingningsviktad</td>
-<td>Laddar allt in i kontexten</td>
-</tr>
-<tr>
-<td><strong>Multiagent</strong></td>
-<td>MCP + REST + leaser + signaler</td>
-<td>API (ingen koordinering)</td>
-<td>Endast inom Letta-körtiden</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Teamroller + delade tillgångar</td>
-<td>Nej</td>
-<td>Endast scopead</td>
-<td>Delad mellan agenter</td>
-<td>Filer per agent</td>
-</tr>
-<tr>
-<td><strong>Ramverksinlåsning</strong></td>
-<td>Ingen (vilken MCP-klient som helst)</td>
-<td>Ingen</td>
-<td>Hög (måste använda Letta)</td>
-<td>Fristående</td>
-<td>Ingen</td>
-<td>Proxy framför varje modellanrop</td>
-<td>Ingen</td>
-<td>Oracle Database</td>
-<td>Ingen</td>
-<td>Format per agent</td>
-</tr>
-<tr>
-<td><strong>Externa beroenden</strong></td>
-<td>Ingen separat databastjänst behövs</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektor-DB</td>
-<td>Flera</td>
-<td>Hanterad molntjänst</td>
-<td>Docker-stack (Core + Hub + Proxy)</td>
-<td>Vektorlager</td>
-<td>Oracle AI Database</td>
-<td>Inga</td>
-<td>Inga</td>
-</tr>
-<tr>
-<td><strong>Minneslivscykel</strong></td>
-<td>4-nivås konsolidering + avklingning + auto-glömning</td>
-<td>Passiv extraktion</td>
-<td>Hanteras av agenten</td>
-<td>Manuellt</td>
-<td>Auto-glömning</td>
-<td>Manuell granskning; automatisk routing på gång</td>
-<td>Inget</td>
-<td>Inte angivet</td>
-<td>Avklingning + konsolidering</td>
-<td>Manuell beskärning</td>
-</tr>
-<tr>
-<td><strong>Tokeneffektivitet</strong></td>
-<td>~1,900 tokens/session ($10/år)</td>
-<td>Varierar efter integration</td>
-<td>Kärnminne i kontexten</td>
-<td>Varierar</td>
-<td>Molnprissättning</td>
-<td>Inte angivet</td>
-<td>Ingen tokenbudget</td>
-<td>LLM-baserad (varierar)</td>
-<td>Varierar</td>
-<td>22K+ tokens vid 240 observationer</td>
-</tr>
-<tr>
-<td><strong>Viewer i realtid</strong></td>
-<td>Ja (port 3113)</td>
-<td>Molnpanel</td>
-<td>Molnpanel</td>
-<td>Webbgränssnitt</td>
-<td>Molnpanel</td>
-<td>Webbgränssnitt för hubben</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Nej</td>
-<td>Nej</td>
-</tr>
-<tr>
-<td><strong>Självhostad</strong></td>
-<td>Ja (standard)</td>
-<td>Valfritt</td>
-<td>Valfritt</td>
-<td>Ja</td>
-<td>Nej (endast moln)</td>
-<td>Ja (Docker)</td>
-<td>Ja</td>
-<td>Ja (Oracle DB)</td>
-<td>Ja</td>
-<td>Ja</td>
-</tr>
-</table>
-
-<sub>Benchmark-notis: det är bara agentmemorys R@5 som är vårt eget uppmätta resultat (LongMemEval-S, reproducerbart från <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Siffrorna för mem0 och Letta är deras publicerade LoCoMo-siffror (ett annat dataset); siffrorna för MemPalace, supermemory, TencentDB (PersonaMem) och oracleagentmemory är leverantörens egna rapporterade påståenden som vi inte har reproducerat oberoende (oracleagentmemorys körning använde GPT-5.5 mot en Oracle AI Database). Visas sida vid sida enbart som fingervisning, inte en head-to-head på identisk data. Stjärnantal är ungefärliga och förändras över tid.</sub>
-
-**Nyare aktörer** värda att känna till, jämförda på djupet i [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | Inriktning |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal kunskapsgraf; starkast publicerade resultat för temporala frågor (LongMemEval 63.8%), men grafen byggs asynkront så färska fakta kan ligga efter |
-| Cognee | 30K | Dokument-till-kunskapsgraf-inmatning, endast Python, byggt för strukturerad entitetsextraktion snarare än sessionsinfångning |
+Uppdaterad jämförelse av produkter och driftsalternativ med källor:
-AgentMemory kombinerar registrering från kodningsagenter, sökbar projekthistorik och en lokal viewer. Du kan arbeta utan nyckel från första starten.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -779,7 +571,7 @@ Fullständig guide: [`integrations/hermes/`](../integrations/hermes/)
Starta minnesservern: `npx -y @agentmemory/agentmemory@latest`
-#### Inbyggda skills via `npx skills add` (50+ agenter)
+#### Inbyggda skills via `npx skills add`
agentmemory levereras med 17 skills i Claude-Code-stilens `<dir>/SKILL.md`-format: 9 anropsbara handlings-skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) och 8 referens-skills som agenten laddar vid behov (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Referens-skillsen innehåller datatabeller genererade från källkoden, så de avviker aldrig. CLI-verktyget [`skills`](https://npmjs.com/package/skills) från vercel-labs installerar dem automatiskt i den anropande agentens inbyggda skill-katalog över 50+ agenter (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf och fler):
@@ -836,7 +628,7 @@ agentmemory-posten är **samma MCP-serverblock** på alla värdar som använder
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` installerar MCP och infångnings-hooks i den delade anpassningskatalogen. Se [Antigravity-konfiguration och begränsningar](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` använder samma MCP- och hook-konfiguration som nuvarande IDE-versioner. Befintliga installationer bör uppdateras med `--force`; se [uppgraderingsanteckningarna](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` skriver konfigurationen på användarnivå. Arbetsyteöverskrivningar läggs i `.kiro/settings/mcp.json` bredvid din kod. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver det vanliga `mcpServers`-blocket. Warp upptäcker också automatiskt skills från `.claude/skills/`; när Claude Code-pluginet är installerat visas agentmemorys 8 skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) naturligt i Warps snabbkommando-palett. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` skriver det vanliga `mcpServers`-blocket. Warp upptäcker också automatiskt skills från `.claude/skills/`; när Claude Code-pluginet är installerat visas agentmemorys 9 skills (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) naturligt i Warps snabbkommando-palett. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` skriver det vanliga `mcpServers`-blocket. VS Code-tilläggsanvändare: klistra in samma block via Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (att föredra) eller `config.json` (äldre) | `agentmemory connect continue` skapar en saknad `config.yaml` eller ändrar befintlig `config.json`. Om YAML-filen finns visas blocket för `mcpServers:` så att kommentarer och ankare bevaras. Continue använder en array, inte ett objekt. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` skriver under `context_servers` (Zeds nyckel, INTE `mcpServers`). Fjärranslutna MCP-servrar kan istället kopplas in via `{"url": "..."}`. |
@@ -933,7 +725,6 @@ Varje AI-kodningsagent levereras med inbyggt minne: Claude Code har `MEMORY.md`,
|---|---|---|
| Underhåll | Manuellt underhållna instruktioner | Indexerad sessionshistorik och minnen |
| Sökning | Laddar allt in i kontexten | BM25 + vektor + graf (endast topp-K) |
-| Tokenkostnad | 22K+ vid 240 observationer | ~1,900 tokens (92% mindre) |
| Mellan agenter | Filer per agent | MCP + REST (vilken agent som helst) |
| Koordinering | Ingen | Leaser, signaler, åtgärder, rutiner |
| Observerbarhet | Läs filer manuellt | Viewer i realtid på :3113 |
@@ -946,7 +737,7 @@ Varje AI-kodningsagent levereras med inbyggt minne: Claude Code har `MEMORY.md`,
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Sammanfogad med Reciprocal Rank Fusion (RRF, k=60) och sessionsdiversifierad (ma
När ett vektorindex är ifyllt använder `mem::search` (bakom `memory_recall`) den hybrida BM25 + vektor-rankaren. Utan embeddings används BM25. `smart-search` kan dessutom sammanfoga strukturella grafmatchningar när grafdata finns, även i nyckelfritt läge. Återkallning av lärdomar körs på ett dedikerat BM25-index i minnet istället för att skanna hela korpuset per fråga. Ersatta minnesversioner exkluderas från varje återkallningsväg; versionskedjan behåller deras historik.
-Vektorer överlever en krasch eller en tvingad avstängning. Vektorindexet sparas i buckets högst var `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minuter). Varje vektor som läggs till eller tas bort däremellan skrivs också direkt till en liten väntande logg i state-lagret, och nästa start spelar upp den igen utan att anropa embedding-leverantören. Varje lyckad sparning tömmer loggen. Dokument som fortfarande saknar en vektor efter uppspelningen bäddas in igen i bakgrunden i omgångar om `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) tills inga återstår, och en backfill som stoppas fortsätter vid nästa start. `/agentmemory/status` och viewern visar storleken på den väntande loggen och backfill-tillståndet. Nyckelfria installationer skriver ingenting.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokeniserar grekiska, kyrilliska, hebreiska, arabiska och accentuerad latin direkt ur lådan. För kinesiska/japanska/koreanska minnen, installera de valfria segmenterarna (`npm install @node-rs/jieba tiny-segmenter`) för att dela upp CJK-sekvenser i ordnivå-tokens; utan dem faller agentmemory mjukt tillbaka till tokenisering av hela sekvenser och skriver ut en engångstips på stderr.
@@ -1047,9 +838,9 @@ Den vanliga npm-installationen inkluderar den valfria körtiden `@huggingface/tr
| Leverantör | Modell | Kostnad | Anteckningar |
|---|---|---|---|
-| **Lokal (rekommenderat att aktivera)** | `all-MiniLM-L6-v2` | Kostnadsfri | På enheten efter den första modellnedladdningen, +8 procentenheter recall jämfört med endast BM25 |
+| **Lokal (rekommenderat att aktivera)** | `all-MiniLM-L6-v2` | Kostnadsfri | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Kostnadsfri nivå | 100+ språk, 768/1536/3072 dimensioner (MRL), 2048-token indata. Ersätter `text-embedding-004` ([utfasad, stängs ner 14 jan 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Högst kvalitet |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Betald | Optimerad för kod |
| Cohere | `embed-english-v3.0` | Kostnadsfri provperiod | Generell användning |
| OpenRouter | Vilken modell som helst | Varierar | Multimodell-proxy |
diff --git a/READMEs/README.th-TH.md b/READMEs/README.th-TH.md
index 93b2dde0..d6752236 100644
--- a/READMEs/README.th-TH.md
+++ b/READMEs/README.th-TH.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="ลด token ลง 92%" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 เครื่องมือ MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 hook อัตโนมัติ" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="ไม่ต้องพึ่ง external DB เลย" height="38" /></picture>
@@ -329,221 +327,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Retrieval Accuracy
-
-**coding-agent-life-v1** (คอร์ปัสภายใน ทำซ้ำได้ใน sandbox)
-
-| Adapter | P@5 | R@5 | Top-5 hit rate | p50 latency |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Top-5 hit rate 100% ที่ **เพดานทางคณิตศาสตร์ของ P@5** สำหรับคอร์ปัสนี้ (0.240 ดูใน scorecard) hybrid ดึงข้อมูล gold session ได้ครบทุกตัว ส่วน grep พลาด 1 ใน 2 ของ gold ในคำค้นที่ครอบคลุมหลายเซสชันตามลำดับเวลา ส่วนที่ดีขึ้นคือ **recall + temporal** ไม่ใช่ precision โดยรวม benchmark นี้มีขนาดเล็กและ gold มีน้อย ส่วน LongMemEval-S ที่ใหญ่กว่าด้านล่างจะแยกความแตกต่างได้ดีกว่า รายละเอียดแยกตามประเภทแบบเต็ม พร้อมหมายเหตุการแก้ไข: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)
-
-**LongMemEval-S** (ICLR 2025, 500 คำถาม)
-
-| System | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token Savings
+ผลการค้นคืนล่าสุด นิยามตัวชี้วัด และขั้นตอนการทดสอบซ้ำ:
-| Approach | Tokens/yr | Cost/yr |
-|---|---|---|
-| Paste full context | 19.5M+ | เป็นไปไม่ได้ (เกิน window) |
-| LLM-summarized | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + local embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding model: `all-MiniLM-L6-v2` (local, ฟรี, ไม่ต้องใช้ API key) รายงานฉบับเต็ม: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md) การเทียบกับคู่แข่ง: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) ครอบคลุม agentmemory เทียบกับ mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo
-
-**ใช้คู่กับ [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), และ [Graphify](https://github.com/safishamsi/graphify)** การทำ index แบบ code-graph, pipeline การ build แบบ multi-agent, และ knowledge graph ที่กว้างขึ้นครอบคลุม docs / PDF / รูปภาพ / วิดีโอ agentmemory จำงานที่ทำไปได้ ส่วนสามโครงการนี้ช่วยเติมเต็ม context layer ส่วนที่เหลือ สูตรการใช้งาน พร้อมตารางเลือกตามคำถาม: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md)
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="เทียบกับคู่แข่ง" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Built-in (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>ประเภท</strong></td>
-<td>Memory engine + MCP server</td>
-<td>Memory layer API</td>
-<td>Full agent runtime</td>
-<td>Personal AI</td>
-<td>Memory API + app</td>
-<td>Team memory hub (LLM proxy)</td>
-<td>Vector memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Memory system</td>
-<td>Static file</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>รายงานเอง</td>
-<td>PersonaMem 76% (รายงานเอง)</td>
-<td>~96.6% (รายงานเอง)</td>
-<td>94.4% (รายงานเอง)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hooks (ไม่ต้องทำเอง)</td>
-<td>เรียก <code>add()</code> เอง</td>
-<td>Agent แก้ไขตัวเอง</td>
-<td>ทำเอง</td>
-<td>ดึงข้อมูลฝั่ง API</td>
-<td>Proxy interception (สลับ base-URL)</td>
-<td>ทำเอง</td>
-<td>ดึงข้อมูลผ่าน API</td>
-<td>ทำเอง</td>
-<td>แก้ไขเอง</td>
-</tr>
-<tr>
-<td><strong>Search</strong></td>
-<td>BM25 + Vector + Graph (RRF fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semantic</td>
-<td>Vector + RAG</td>
-<td>4 asset types (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vector-only</td>
-<td>Vector + semantic</td>
-<td>Decay-weighted</td>
-<td>โหลดทุกอย่างเข้า context</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + leases + signals</td>
-<td>API (ไม่มีการประสานงาน)</td>
-<td>ภายใน Letta runtime เท่านั้น</td>
-<td>ไม่มี</td>
-<td>ไม่มี</td>
-<td>Team roles + shared assets</td>
-<td>ไม่มี</td>
-<td>จำกัดเฉพาะ scope</td>
-<td>Multi-agent shared</td>
-<td>ไฟล์แยกตาม agent</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>ไม่มี (MCP client ใดก็ได้)</td>
-<td>ไม่มี</td>
-<td>สูง (ต้องใช้ Letta)</td>
-<td>Standalone</td>
-<td>ไม่มี</td>
-<td>Proxy ครอบทุก model call</td>
-<td>ไม่มี</td>
-<td>Oracle Database</td>
-<td>ไม่มี</td>
-<td>รูปแบบแยกตาม agent</td>
-</tr>
-<tr>
-<td><strong>External deps</strong></td>
-<td>ไม่ต้องมีบริการฐานข้อมูลแยก</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>หลายตัว</td>
-<td>Managed cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>ไม่มี</td>
-<td>ไม่มี</td>
-</tr>
-<tr>
-<td><strong>Memory lifecycle</strong></td>
-<td>4-tier consolidation + decay + auto-forget</td>
-<td>ดึงข้อมูลแบบ passive</td>
-<td>Agent จัดการเอง</td>
-<td>ทำเอง</td>
-<td>Auto-forget</td>
-<td>ตรวจสอบเอง; auto-routing กำลังพัฒนา</td>
-<td>ไม่มี</td>
-<td>ไม่ระบุ</td>
-<td>Decay + consolidation</td>
-<td>ตัดทิ้งเอง</td>
-</tr>
-<tr>
-<td><strong>Token efficiency</strong></td>
-<td>~1,900 token/เซสชัน ($10/ปี)</td>
-<td>แตกต่างกันตามการเชื่อมต่อ</td>
-<td>Core memory อยู่ใน context</td>
-<td>แตกต่างกัน</td>
-<td>ราคาแบบ cloud</td>
-<td>ไม่ระบุ</td>
-<td>ไม่มี token budget</td>
-<td>ขึ้นกับ LLM (แตกต่างกัน)</td>
-<td>แตกต่างกัน</td>
-<td>22K+ token ที่ 240 observation</td>
-</tr>
-<tr>
-<td><strong>Real-time viewer</strong></td>
-<td>มี (พอร์ต 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>ไม่มี</td>
-<td>ไม่มี</td>
-<td>ไม่มี</td>
-<td>ไม่มี</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>มี (default)</td>
-<td>เป็นตัวเลือก</td>
-<td>เป็นตัวเลือก</td>
-<td>มี</td>
-<td>ไม่มี (cloud เท่านั้น)</td>
-<td>มี (Docker)</td>
-<td>มี</td>
-<td>มี (Oracle DB)</td>
-<td>มี</td>
-<td>มี</td>
-</tr>
-</table>
-
-<sub>หมายเหตุเรื่อง benchmark: มีเพียงค่า R@5 ของ agentmemory เท่านั้นที่เป็นผลวัดของเราเอง (LongMemEval-S ทำซ้ำได้จาก <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>) ตัวเลขของ mem0 และ Letta เป็นตัวเลข LoCoMo ที่พวกเขาเผยแพร่เอง (คนละ dataset) ส่วนตัวเลขของ MemPalace, supermemory, TencentDB (PersonaMem) และ oracleagentmemory เป็นข้อเรียกร้องที่ผู้ผลิตรายงานเอง ซึ่งเรายังไม่ได้ทำซ้ำเพื่อตรวจสอบอย่างเป็นอิสระ (การรันของ oracleagentmemory ใช้ GPT-5.5 กับ Oracle AI Database) แสดงเทียบกันเพื่อให้เห็นภาพกว้าง ๆ เท่านั้น ไม่ใช่การเทียบแบบตัวต่อตัวบนข้อมูลชุดเดียวกัน จำนวน star เป็นค่าประมาณและเปลี่ยนแปลงไปตามเวลา</sub>
-
-**ผู้เล่นใหม่** ที่ควรรู้จัก เทียบรายละเอียดได้ใน [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| System | ⭐ | มุมมอง |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal knowledge graph; ผลลัพธ์ temporal-query ที่เผยแพร่แล้วแข็งแกร่งที่สุด (LongMemEval 63.8%) แต่ graph สร้างแบบ asynchronous จึงอาจตามข้อมูลใหม่ไม่ทัน |
-| Cognee | 30K | การนำ document เข้าสู่ knowledge graph, รองรับแค่ Python, สร้างมาเพื่อดึง structured entity มากกว่าการ capture เซสชัน |
+การเปรียบเทียบผลิตภัณฑ์และตัวเลือกการติดตั้งล่าสุดพร้อมแหล่งอ้างอิง:
-AgentMemory รวมการบันทึกเซสชัน ประวัติโปรเจ็กต์ที่ค้นหาได้ และตัวดูข้อมูลในเครื่อง พร้อมใช้งานแบบไม่ต้องมีคีย์ตั้งแต่ครั้งแรก
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -770,7 +564,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
เริ่ม memory server: `npx -y @agentmemory/agentmemory@latest`
-#### Native skills via `npx skills add` (50+ agents)
+#### Native skills via `npx skills add`
agentmemory ship skill มาทั้งหมด 17 ตัว ในรูปแบบ `<dir>/SKILL.md` สไตล์ Claude Code: action skill ที่เรียกใช้ได้ 9 ตัว (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) และ reference skill อีก 8 ตัวที่ agent โหลดตามต้องการ (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`) reference skill พวกนี้มีตารางข้อมูลที่สร้างจาก source โดยตรง จึงไม่มีทางเพี้ยนไปจากของจริง CLI [`skills`](https://npmjs.com/package/skills) ของ vercel-labs จะติดตั้งให้อัตโนมัติเข้าไปใน native skill directory ของ agent ที่เรียกมันขึ้นมา ครอบคลุม agent กว่า 50 ตัว (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, และอีกมากมาย):
@@ -827,7 +621,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` จะติดตั้ง MCP และ capture hook ไว้ใน shared customization directory ดู [การตั้งค่าและข้อจำกัดของ Antigravity](../docs/plugins/antigravity.md) |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` ใช้ config ของ MCP และ hook แบบเดียวกับ IDE เวอร์ชันปัจจุบัน การติดตั้งที่มีอยู่แล้วควร refresh ด้วย `--force`; ดู [upgrade notes](../docs/plugins/antigravity.md) |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` เขียน config ระดับ user การ override ระดับ workspace ไปอยู่ที่ `.kiro/settings/mcp.json` ข้าง ๆ โค้ดของคุณ |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` เขียน `mcpServers` block แบบมาตรฐาน Warp ยัง auto-discover skill จาก `.claude/skills/` ด้วย เมื่อติดตั้ง Claude Code plugin แล้ว skill ทั้ง 8 ตัวของ agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) จะปรากฏใน slash-command palette ของ Warp โดยตรง |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` เขียน `mcpServers` block แบบมาตรฐาน Warp ยัง auto-discover skill จาก `.claude/skills/` ด้วย เมื่อติดตั้ง Claude Code plugin แล้ว skill ทั้ง 9 ตัวของ agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) จะปรากฏใน slash-command palette ของ Warp โดยตรง |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` เขียน `mcpServers` block แบบมาตรฐาน ผู้ใช้ VS Code extension: paste block เดียวกันผ่าน Cline Settings → MCP Servers → Edit JSON |
| **Continue.dev** | `~/.continue/config.yaml` (แนะนำ) หรือ `config.json` (รุ่นเก่า) | `agentmemory connect continue` สร้าง `config.yaml` เมื่อยังไม่มีคอนฟิก หรืออัปเดต `config.json` ที่มีอยู่ สำหรับ YAML เดิมจะเก็บ comment และ anchor ไว้โดยไม่แก้ไฟล์ แต่แสดงบล็อกให้วางใต้ `mcpServers:` ซึ่ง Continue ใช้รูปแบบ array |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` เขียนไว้ใต้ `context_servers` (key ของ Zed ไม่ใช่ `mcpServers`) MCP server แบบ remote เชื่อมต่อผ่าน `{"url": "..."}` ได้เช่นกัน |
@@ -924,7 +718,6 @@ coding agent ที่ใช้ AI ทุกตัวมีหน่วยคว
|---|---|---|
| การดูแล | คำสั่งที่จัดการด้วยตนเอง | ประวัติเซสชันและความจำที่จัดทำดัชนี |
| Search | โหลดทุกอย่างเข้า context | BM25 + vector + graph (top-K เท่านั้น) |
-| Token cost | 22K+ ที่ 240 observation | ~1,900 token (ลดลง 92%) |
| Cross-agent | ไฟล์แยกตาม agent | MCP + REST (agent ใดก็ได้) |
| Coordination | ไม่มี | Leases, signals, actions, routines |
| Observability | อ่านไฟล์ด้วยตัวเอง | Real-time viewer บน :3113 |
@@ -937,7 +730,7 @@ coding agent ที่ใช้ AI ทุกตัวมีหน่วยคว
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1022,7 +815,7 @@ SessionStart hook fires
เมื่อ vector index มีข้อมูลอยู่ `mem::search` (อยู่หลัง `memory_recall`) จะใช้ hybrid ranker แบบ BM25 + vector โดยไม่มี embedding มันจะใช้ BM25 `smart-search` สามารถผสาน structural graph match เพิ่มเติมได้เมื่อมีข้อมูล graph อยู่ รวมถึงในโหมด keyless ด้วย การ recall lesson รันบน BM25 index ในหน่วยความจำเฉพาะของมันเอง แทนการสแกนทั้งคอร์ปัสทุกครั้งที่ค้นหา memory version ที่ถูกแทนที่แล้วจะถูกตัดออกจากทุกเส้นทาง recall; version chain ยังคง history ไว้
-Vector รอดจากการ crash หรือ force-kill ได้ vector index จะถูกบันทึกเป็น bucket อย่างมากที่สุดทุก ๆ `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 นาที) vector ทุกตัวที่เพิ่มหรือลบในระหว่างนั้นจะถูกเขียนลง pending log เล็ก ๆ ใน state store ทันทีด้วย และการเริ่มครั้งถัดไปจะ replay มันโดยไม่ต้องเรียก embedding provider การบันทึกที่สำเร็จแต่ละครั้งจะล้าง log ให้ว่าง document ที่ยังไม่มี vector หลัง replay จะถูก re-embed ในพื้นหลังเป็น batch ละ `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) จนกว่าจะไม่เหลือสักตัว และ backfill ที่ถูกหยุดจะทำต่อในการเริ่มครั้งถัดไป `/agentmemory/status` และ viewer จะแสดงขนาด pending log และสถานะ backfill การติดตั้งแบบ keyless จะไม่เขียนอะไรเลย
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenize ภาษากรีก, ซีริลลิก, ฮีบรู, อารบิก, และละตินแบบมี accent ได้ทันทีโดยไม่ต้องตั้งค่าเพิ่ม สำหรับหน่วยความจำภาษาจีน / ญี่ปุ่น / เกาหลี ให้ติดตั้ง segmenter แบบ optional (`npm install @node-rs/jieba tiny-segmenter`) เพื่อแยกข้อความ CJK เป็น token ระดับคำ; ถ้าไม่มี agentmemory จะ soft-fallback ไปที่การ tokenize ทั้ง run และพิมพ์คำเตือนครั้งเดียวออกที่ stderr
@@ -1038,9 +831,9 @@ EMBEDDING_PROVIDER=local
| Provider | Model | Cost | Notes |
|---|---|---|---|
-| **Local (แนะนำให้เปิดเอง)** | `all-MiniLM-L6-v2` | ฟรี | ทำงานในเครื่องหลังดาวน์โหลดโมเดลครั้งแรก, recall เพิ่มขึ้น +8pp เทียบกับ BM25-only |
+| **Local (แนะนำให้เปิดเอง)** | `all-MiniLM-L6-v2` | ฟรี | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | มี free tier | รองรับ 100+ ภาษา, 768/1536/3072 มิติ (MRL), input 2048 token แทนที่ `text-embedding-004` ([deprecated, ปิดให้บริการ 14 ม.ค. 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | คุณภาพสูงที่สุด |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | เสียเงิน | ปรับให้เหมาะกับโค้ด |
| Cohere | `embed-english-v3.0` | มี free trial | ใช้งานทั่วไป |
| OpenRouter | โมเดลใดก็ได้ | ขึ้นกับโมเดล | Proxy แบบ multi-model |
diff --git a/READMEs/README.tl-PH.md b/READMEs/README.tl-PH.md
index 4c8e2825..acc9b243 100644
--- a/READMEs/README.tl-PH.md
+++ b/READMEs/README.tl-PH.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% mas kaunting token" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tool" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 awtomatikong hook" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 panlabas na DB" height="38" /></picture>
@@ -331,223 +329,17 @@ Pinakabagong release notes: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Katumpakan ng Retrieval
-
-**coding-agent-life-v1** (in-house corpus, reproducible sa sandbox)
-
-| Adapter | P@5 | R@5 | Top-5 hit rate | p50 latency |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% top-5 hit rate sa **P@5 math ceiling** para sa corpus na ito (0.240, tingnan ang scorecard). Nakukuha ng hybrid ang bawat gold session; nawawala sa grep ang 1 sa 2 gold sa multi-session temporal query. Ang lift ay sa **recall + temporal**, hindi sa aggregate precision. Maliit at gold-sparse ang benchmark na ito; mas mahusay na nagkakaiba ang mas malaking LongMemEval-S sa ibaba. Buong per-type breakdown + correction note: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 tanong)
+Kasalukuyang resulta ng paghahanap, kahulugan ng mga sukatan, at mga hakbang sa pag-ulit ng pagsubok:
-| Sistema | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Pagtitipid sa Token
-
-| Pamamaraan | Tokens/yr | Cost/yr |
-|---|---|---|
-| I-paste ang buong context | 19.5M+ | Imposible (lumalagpas sa window) |
-| Pinaikli ng LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + local embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding model: `all-MiniLM-L6-v2` (lokal, libre, walang API key). Kumpletong mga ulat: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Paghambing sa kompetisyon: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) na sumasaklaw sa agentmemory laban sa mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**I-reproduce nang lokal:** [`eval/README.md`](../eval/README.md), isang adapter-pluggable harness para sa LongMemEval `_s` (public 500-Q) + `coding-agent-life-v1` (in-house na 15-session corpus). Nag-iiskor ang mga grep / vector / agentmemory adapter nang tabi-tabi, NDJSON output, dumadapo ang mga published scorecard sa [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Umaangkop sa [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), at [Graphify](https://github.com/safishamsi/graphify).** Code-graph indexing, multi-agent build pipeline, at mas malawak na knowledge graph sa mga docs / PDF / imahe / video. Naaalala ng agentmemory ang ginawang trabaho; pinapaliwanag naman ng tatlong project na ito ang iba pang bahagi ng context layer. Mga recipe + question-routing table: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="VS Kompetisyon" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Built-in (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Uri</strong></td>
-<td>Memory engine + MCP server</td>
-<td>Memory layer API</td>
-<td>Buong agent runtime</td>
-<td>Personal AI</td>
-<td>Memory API + app</td>
-<td>Team memory hub (LLM proxy)</td>
-<td>Vector memory (OSS)</td>
-<td>Memory engine (Oracle DB)</td>
-<td>Sistema ng memory</td>
-<td>Static file</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Sariling ulat</td>
-<td>PersonaMem 76% (sariling ulat)</td>
-<td>~96.6% (sariling ulat)</td>
-<td>94.4% (sariling ulat)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hook (walang manu-manong pagsisikap)</td>
-<td>Manual na <code>add()</code> call</td>
-<td>Sariling pag-edit ng agent</td>
-<td>Manual</td>
-<td>API-side extraction</td>
-<td>Proxy interception (pagpalit ng base-URL)</td>
-<td>Manual</td>
-<td>API extraction</td>
-<td>Manual</td>
-<td>Manual na pag-edit</td>
-</tr>
-<tr>
-<td><strong>Paghahanap</strong></td>
-<td>BM25 + Vector + Graph (RRF fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (archival)</td>
-<td>Semantic</td>
-<td>Vector + RAG</td>
-<td>4 uri ng asset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Vector lamang</td>
-<td>Vector + semantic</td>
-<td>Decay-weighted</td>
-<td>Nilo-load ang lahat sa context</td>
-</tr>
-<tr>
-<td><strong>Multi-agent</strong></td>
-<td>MCP + REST + lease + signal</td>
-<td>API (walang coordination)</td>
-<td>Sa loob lamang ng Letta runtime</td>
-<td>Hindi</td>
-<td>Hindi</td>
-<td>Mga tungkulin sa team + shared asset</td>
-<td>Hindi</td>
-<td>Scoped lamang</td>
-<td>Shared sa multi-agent</td>
-<td>Per-agent na file</td>
-</tr>
-<tr>
-<td><strong>Framework lock-in</strong></td>
-<td>Wala (kahit anong MCP client)</td>
-<td>Wala</td>
-<td>Mataas (kailangang gamitin ang Letta)</td>
-<td>Standalone</td>
-<td>Wala</td>
-<td>Proxy sa harap ng bawat model call</td>
-<td>Wala</td>
-<td>Oracle Database</td>
-<td>Wala</td>
-<td>Per-agent na format</td>
-</tr>
-<tr>
-<td><strong>Panlabas na deps</strong></td>
-<td>Hindi kailangan ng hiwalay na database service</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>Marami</td>
-<td>Managed cloud</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Wala</td>
-<td>Wala</td>
-</tr>
-<tr>
-<td><strong>Memory lifecycle</strong></td>
-<td>4-tier consolidation + decay + auto-forget</td>
-<td>Passive extraction</td>
-<td>Pinamamahalaan ng agent</td>
-<td>Manual</td>
-<td>Auto-forget</td>
-<td>Manual review; isinasagawa pa ang auto-routing</td>
-<td>Wala</td>
-<td>Hindi nakasaad</td>
-<td>Decay + consolidation</td>
-<td>Manual na pruning</td>
-</tr>
-<tr>
-<td><strong>Kahusayan ng Token</strong></td>
-<td>~1,900 tokens/session ($10/yr)</td>
-<td>Depende sa integration</td>
-<td>Core memory sa context</td>
-<td>Depende</td>
-<td>Cloud pricing</td>
-<td>Hindi nakasaad</td>
-<td>Walang token budget</td>
-<td>LLM-backed (depende)</td>
-<td>Depende</td>
-<td>22K+ tokens sa 240 obs</td>
-</tr>
-<tr>
-<td><strong>Real-time viewer</strong></td>
-<td>Oo (port 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>Hindi</td>
-<td>Hindi</td>
-<td>Hindi</td>
-<td>Hindi</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Oo (default)</td>
-<td>Opsyonal</td>
-<td>Opsyonal</td>
-<td>Oo</td>
-<td>Hindi (cloud-only)</td>
-<td>Oo (Docker)</td>
-<td>Oo</td>
-<td>Oo (Oracle DB)</td>
-<td>Oo</td>
-<td>Oo</td>
-</tr>
-</table>
-
-<sub>Tala sa benchmark: ang R@5 lamang ng agentmemory ang aming sariling nasukat na resulta (LongMemEval-S, reproducible mula sa <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Ang mga figure ng mem0 at Letta ay ang kanilang mga pinublikang numero sa LoCoMo (ibang dataset); ang mga figure ng MemPalace, supermemory, TencentDB (PersonaMem), at oracleagentmemory ay mga claim na self-reported ng vendor na hindi namin independent na-reproduce (ang run ng oracleagentmemory ay gumamit ng GPT-5.5 laban sa isang Oracle AI Database). Ipinapakita nang tabi-tabi para sa ballpark estimate lamang, hindi isang head-to-head sa parehong data. Ang bilang ng star ay tinatantya at nagbabago sa paglipas ng panahon.</sub>
-
-**Mga bagong entrant** na dapat malaman, inihambing nang detalyado sa [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Sistema | ⭐ | Anggulo |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal knowledge graph; ang pinakamalakas na pinublikang resulta sa temporal-query (LongMemEval 63.8%), pero bumubuo ang graph nang asynchronous kaya maaaring mahuli ang mga bagong fact |
-| Cognee | 30K | Document-to-knowledge-graph ingestion, Python lamang, ginawa para sa structured entity extraction sa halip na session capture |
+Na-update na paghahambing ng mga produkto at paraan ng deployment, kasama ang mga sanggunian:
-Pinagsasama ng AgentMemory ang session capture, searchable na project history, at local viewer, na magagamit nang walang key mula sa unang pagpapatakbo.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Kumpletong gabay: [`integrations/hermes/`](../integrations/hermes/)
Simulan ang memory server: `npx -y @agentmemory/agentmemory@latest`
-#### Native na skill gamit ang `npx skills add` (50+ agent)
+#### Native na skill gamit ang `npx skills add`
Nag-ship ang agentmemory ng 17 skill sa Claude-Code-style na `<dir>/SKILL.md` format: 9 invocable na action skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) at 8 reference skill na nilo-load ng agent kapag kailangan (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). May dala ang mga reference skill na data table na nabuo mula sa source, kaya hindi ito kailanman nadrift. Ang [`skills`](https://npmjs.com/package/skills) CLI ng vercel-labs ay awtomatikong nag-install nito sa native skill directory ng tumatawag na agent sa 50+ agent (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, at iba pa):
@@ -837,7 +629,7 @@ Ang entry ng agentmemory ay ang **parehong MCP server block** sa bawat host na g
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | Ini-install ng `agentmemory connect antigravity --with-hooks` ang MCP at capture hooks sa shared customization directory. Tingnan ang [setup at limitasyon ng Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | Ginagamit ng `agentmemory connect antigravity-cli --with-hooks` ang parehong MCP at hook configuration ng kasalukuyang mga bersyon ng IDE. Dapat i-refresh ng mga umiiral na installation gamit ang `--force`; tingnan ang [upgrade notes](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | Isinusulat ng `agentmemory connect kiro` ang user-level config. Ang workspace override ay sa `.kiro/settings/mcp.json`, katabi ng iyong code. |
-| **Warp** | `~/.warp/.mcp.json` | Isinusulat ng `agentmemory connect warp` ang standard na `mcpServers` block. Awtomatikong nadidiskubre din ng Warp ang mga skill mula sa `.claude/skills/`; kapag naka-install na ang Claude Code plugin, lumalabas nang native ang 8 agentmemory skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) sa slash-command palette ng Warp. |
+| **Warp** | `~/.warp/.mcp.json` | Isinusulat ng `agentmemory connect warp` ang standard na `mcpServers` block. Awtomatikong nadidiskubre din ng Warp ang mga skill mula sa `.claude/skills/`; kapag naka-install na ang Claude Code plugin, lumalabas nang native ang 9 agentmemory skill (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) sa slash-command palette ng Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | Isinusulat ng `agentmemory connect cline` ang standard na `mcpServers` block. Para sa VS Code extension user: i-paste ang parehong block sa Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (preferred) o `config.json` (legacy) | Gumagawa ang `agentmemory connect continue` ng `config.yaml` kung wala pang configuration o ina-update ang kasalukuyang `config.json`. Hindi nito binabago ang umiiral na YAML upang mapanatili ang comments at anchors; ipinapakita ang block na ilalagay sa ilalim ng `mcpServers:`. Array ang gamit ng Continue para sa `mcpServers`. |
| **Zed** | `~/.config/zed/settings.json` | Isinusulat ng `agentmemory connect zed` sa ilalim ng `context_servers` (ang key ng Zed, HINDI `mcpServers`). Maaaring i-wire ang remote MCP server gamit ang `{"url": "..."}` sa halip. |
@@ -933,7 +725,6 @@ Nag-ship ang bawat AI coding agent na may built-in memory: may `MEMORY.md` ang C
|---|---|---|
| Pagpapanatili | Manu-manong inayos na instructions | Naka-index na session history at memory |
| Paghahanap | Nilo-load ang lahat sa context | BM25 + vector + graph (top-K lamang) |
-| Gastos sa Token | 22K+ sa 240 obserbasyon | ~1,900 token (92% mas kaunti) |
| Cross-agent | Per-agent na file | MCP + REST (kahit anong agent) |
| Koordinasyon | Wala | Lease, signal, action, routine |
| Observability | Basahin ang mga file nang manu-mano | Real-time viewer sa :3113 |
@@ -946,7 +737,7 @@ Nag-ship ang bawat AI coding agent na may built-in memory: may `MEMORY.md` ang C
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Pinagsasama gamit ang Reciprocal Rank Fusion (RRF, k=60) at session-diversified
Kapag napupuno ang isang vector index, ginagamit ng `mem::search` (sa likod ng `memory_recall`) ang hybrid na BM25 + vector ranker. Kapag walang embeddings, ginagamit nito ang BM25. Maaari pang pagsamahin ng `smart-search` ang structural graph matches kapag umiiral ang graph data, kasama na sa keyless mode. Tumatakbo ang lesson recall sa isang dedikadong in-memory BM25 index sa halip na i-scan ang buong corpus kada query. Hindi kasama sa bawat recall path ang mga superseded na bersyon ng memory; pinapanatili ng version chain ang kanilang history.
-Nakaligtas ang mga vector sa isang crash o force-kill. Nasa-save ang vector index sa mga bucket sa pinakamadalas na bawat `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 minuto). Bawat vector na idinagdag o inalis sa gitna ay agad na isinusulat din sa isang maliit na pending log sa state store, at ni-replay ito ng susunod na start nang hindi tinawag ang embedding provider. Ang bawat matagumpay na save ay nagwawalang-laman sa log. Ang mga documento na wala pa ring vector pagkatapos ng replay ay re-embedded sa background sa mga batch ng `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) hanggang maubos ang lahat, at ang backfill na natigil ay magpapatuloy sa susunod na start. Ipinapakita ng `/agentmemory/status` at ng viewer ang laki ng pending log at ang estado ng backfill. Walang isinusulat ang mga keyless install.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
Tinotokenize ng BM25 ang Griyego, Cyrillic, Hebreo, Arabic, at accented Latin nang out of the box. Para sa mga memory sa Chinese / Japanese / Korean, i-install ang opsyonal na segmenter (`npm install @node-rs/jieba tiny-segmenter`) para hatiin ang mga CJK run sa word-level token; kung wala ito, soft-falls ang agentmemory sa whole-run tokenization at nagprint ng one-time hint sa stderr.
@@ -1047,9 +838,9 @@ Kasama sa normal na npm install ang opsyonal na `@huggingface/transformers` runt
| Provider | Model | Gastos | Mga Tala |
|---|---|---|---|
-| **Local (inirerekomendang opt-in)** | `all-MiniLM-L6-v2` | Libre | On-device pagkatapos ng unang pag-download ng model, +8pp recall kaysa sa BM25-only |
+| **Local (inirerekomendang opt-in)** | `all-MiniLM-L6-v2` | Libre | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free tier | 100+ wika, 768/1536/3072 dims (MRL), 2048-token input. Pumalit sa `text-embedding-004` ([deprecated, magsasara noong Jan 14, 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Pinakamataas na kalidad |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Bayad | Na-optimize para sa code |
| Cohere | `embed-english-v3.0` | Free trial | Pangkalahatang gamit |
| OpenRouter | Kahit anong model | Depende | Multi-model proxy |
diff --git a/READMEs/README.tr-TR.md b/READMEs/README.tr-TR.md
index 014dd3f7..0f8aa10d 100644
--- a/READMEs/README.tr-TR.md
+++ b/READMEs/README.tr-TR.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -333,223 +331,17 @@ En son sürüm notları: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Geri Getirme Doğruluğu
-
-**coding-agent-life-v1** (şirket içi (in-house) corpus, sandbox'ta yeniden üretilebilir)
-
-| Adaptör | P@5 | R@5 | İlk 5'te bulma oranı | p50 gecikme |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Bu corpus için **P@5 matematiksel tavanında** (0.240, scorecard'a bakın) %100 ilk-5 bulma oranı. Hybrid, her gold oturumu getirir; grep, çok-oturumlu temporal sorguda 2 gold'dan 1'ini kaçırır. Kazanım **recall + temporal**'dir, toplam precision değil. Bu benchmark küçüktür ve gold açısından seyrektir; aşağıdaki daha büyük LongMemEval-S daha iyi ayrıştırır. Tam tür bazlı döküm + düzeltme notu: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 soru)
+Güncel bilgi getirme sonuçları, metrik tanımları ve yeniden üretme adımları:
-| Sistem | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| Yalnızca BM25 fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token Tasarrufu
-
-| Yaklaşım | Token/yıl | Maliyet/yıl |
-|---|---|---|
-| Tam bağlamı yapıştır | 19.5M+ | İmkansız (pencereyi aşar) |
-| LLM ile özetlenmiş | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + yerel embedding'ler | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding modeli: `all-MiniLM-L6-v2` (yerel, ücretsiz, API anahtarı gerektirmez). Tam raporlar: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Rakip karşılaştırması: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) — agentmemory'yi mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo ile karşılaştırır.
-
-**Yerelde yeniden üretin:** [`eval/README.md`](../eval/README.md), LongMemEval `_s` (herkese açık 500 soru) + `coding-agent-life-v1` (şirket içi 15-oturumluk corpus) için adaptör-takılabilir bir harness. Grep / vektör / agentmemory adaptörleri yan yana puanlanır, NDJSON çıktısı verir, yayımlanan scorecard'lar [`docs/benchmarks/`](../docs/benchmarks/) içine düşer.
-
-**[codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) ve [Graphify](https://github.com/safishamsi/graphify) ile eşleşir.** Kod-grafı indeksleme, çok-ajanlı build pipeline'ları ve doküman / PDF / görüntü / video genelinde daha geniş bilgi grafları. agentmemory yapılan işi hatırlar; bu üç proje, bağlam katmanının gerisini aydınlatır. Tarifler + soru-yönlendirme tablosu: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Yerleşik (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Tür</strong></td>
-<td>Bellek engine'i + MCP sunucusu</td>
-<td>Bellek katmanı API'si</td>
-<td>Tam ajan runtime'ı</td>
-<td>Kişisel AI</td>
-<td>Bellek API'si + uygulama</td>
-<td>Takım belleği hub'ı (LLM proxy)</td>
-<td>Vektör belleği (OSS)</td>
-<td>Bellek engine'i (Oracle DB)</td>
-<td>Bellek sistemi</td>
-<td>Statik dosya</td>
-</tr>
-<tr>
-<td><strong>Geri Getirme R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Kendi beyanı</td>
-<td>PersonaMem 76% (kendi beyanı)</td>
-<td>~96.6% (kendi beyanı)</td>
-<td>94.4% (kendi beyanı)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Otomatik yakalama</strong></td>
-<td>12 hook (sıfır manuel çaba)</td>
-<td>Manuel <code>add()</code> çağrıları</td>
-<td>Ajan kendi kendine düzenler</td>
-<td>Manuel</td>
-<td>API tarafında çıkarım</td>
-<td>Proxy ile yakalama (base-URL değişimi)</td>
-<td>Manuel</td>
-<td>API çıkarımı</td>
-<td>Manuel</td>
-<td>Manuel düzenleme</td>
-</tr>
-<tr>
-<td><strong>Arama</strong></td>
-<td>BM25 + Vektör + Graf (RRF füzyonu)</td>
-<td>Vektör + Graf</td>
-<td>Vektör (arşivsel)</td>
-<td>Semantik</td>
-<td>Vektör + RAG</td>
-<td>4 varlık türü (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Yalnızca vektör</td>
-<td>Vektör + semantik</td>
-<td>Çürüme ağırlıklı (decay-weighted)</td>
-<td>Her şeyi bağlama yükler</td>
-</tr>
-<tr>
-<td><strong>Çoklu ajan</strong></td>
-<td>MCP + REST + lease'ler + signal'ler</td>
-<td>API (koordinasyon yok)</td>
-<td>Yalnızca Letta runtime'ı içinde</td>
-<td>Hayır</td>
-<td>Hayır</td>
-<td>Takım rolleri + paylaşılan varlıklar</td>
-<td>Hayır</td>
-<td>Yalnızca scope'lu</td>
-<td>Çoklu ajan arasında paylaşılan</td>
-<td>Ajan başına dosyalar</td>
-</tr>
-<tr>
-<td><strong>Framework bağımlılığı (lock-in)</strong></td>
-<td>Yok (herhangi bir MCP istemcisi)</td>
-<td>Yok</td>
-<td>Yüksek (Letta kullanmak zorunlu)</td>
-<td>Bağımsız</td>
-<td>Yok</td>
-<td>Proxy her model çağrısının önünde</td>
-<td>Yok</td>
-<td>Oracle Database</td>
-<td>Yok</td>
-<td>Ajan başına format</td>
-</tr>
-<tr>
-<td><strong>Harici bağımlılıklar</strong></td>
-<td>Ayrı bir veritabanı hizmeti gerektirmez</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vektör DB</td>
-<td>Birden fazla</td>
-<td>Yönetilen bulut</td>
-<td>Docker stack'i (Core + Hub + Proxy)</td>
-<td>Vektör deposu</td>
-<td>Oracle AI Database</td>
-<td>Yok</td>
-<td>Yok</td>
-</tr>
-<tr>
-<td><strong>Bellek yaşam döngüsü</strong></td>
-<td>4 katmanlı konsolidasyon + çürüme + otomatik unutma</td>
-<td>Pasif çıkarım</td>
-<td>Ajan tarafından yönetilen</td>
-<td>Manuel</td>
-<td>Otomatik unutma</td>
-<td>Manuel inceleme; otomatik yönlendirme geliştiriliyor</td>
-<td>Yok</td>
-<td>Belirtilmemiş</td>
-<td>Çürüme + konsolidasyon</td>
-<td>Manuel budama</td>
-</tr>
-<tr>
-<td><strong>Token verimliliği</strong></td>
-<td>~1,900 token/oturum ($10/yıl)</td>
-<td>Entegrasyona göre değişir</td>
-<td>Çekirdek bellek bağlamda</td>
-<td>Değişir</td>
-<td>Bulut fiyatlandırması</td>
-<td>Belirtilmemiş</td>
-<td>Token bütçesi yok</td>
-<td>LLM destekli (değişir)</td>
-<td>Değişir</td>
-<td>240 gözlemde 22K+ token</td>
-</tr>
-<tr>
-<td><strong>Gerçek zamanlı görüntüleyici</strong></td>
-<td>Evet (port 3113)</td>
-<td>Bulut dashboard'u</td>
-<td>Bulut dashboard'u</td>
-<td>Web arayüzü</td>
-<td>Bulut dashboard'u</td>
-<td>Hub web arayüzü</td>
-<td>Hayır</td>
-<td>Hayır</td>
-<td>Hayır</td>
-<td>Hayır</td>
-</tr>
-<tr>
-<td><strong>Kendi sunucunda barındırma</strong></td>
-<td>Evet (varsayılan)</td>
-<td>Opsiyonel</td>
-<td>Opsiyonel</td>
-<td>Evet</td>
-<td>Hayır (yalnızca bulut)</td>
-<td>Evet (Docker)</td>
-<td>Evet</td>
-<td>Evet (Oracle DB)</td>
-<td>Evet</td>
-<td>Evet</td>
-</tr>
-</table>
-
-<sub>Benchmark notu: yalnızca agentmemory'nin R@5'i bizim kendi ölçtüğümüz sonuçtur (LongMemEval-S, <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>'dan yeniden üretilebilir). mem0 ve Letta rakamları, onların yayımladığı LoCoMo sayılarıdır (farklı bir veri kümesi); MemPalace, supermemory, TencentDB (PersonaMem) ve oracleagentmemory rakamları, bağımsız olarak yeniden üretmediğimiz, satıcı tarafından kendi beyan edilen iddialardır (oracleagentmemory'nin çalıştırması bir Oracle AI Database'e karşı GPT-5.5 kullandı). Yalnızca kabaca karşılaştırma için yan yana gösterilmiştir, aynı veri üzerinde doğrudan bir karşılaştırma değildir. Yıldız sayıları yaklaşıktır ve zamanla değişir.</sub>
-
-**Bilinmesi gereken daha yeni oyuncular**, [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) içinde derinlemesine karşılaştırılmıştır:
-
-| Sistem | ⭐ | Yaklaşım |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal bilgi grafı; yayımlanmış en güçlü temporal-sorgu sonuçları (LongMemEval 63.8%), ancak graf asenkron olarak oluşturulur, bu yüzden yeni bilgiler gecikebilir |
-| Cognee | 30K | Doküman-to-bilgi-grafı alımı, yalnızca Python, oturum yakalamadan ziyade yapılandırılmış entity çıkarımı için inşa edilmiştir |
+Kaynaklarıyla birlikte ürünlerin ve dağıtım seçeneklerinin güncel karşılaştırması:
-AgentMemory oturum kaydını, aranabilir proje geçmişini ve yerel görüntüleyiciyi birleştirir; anahtarsız çalışma ilk başlatmadan itibaren kullanılabilir.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Tam kılavuz: [`integrations/hermes/`](../integrations/hermes/)
Bellek sunucusunu başlatın: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add` ile yerel skill'ler (50+ ajan)
+#### `npx skills add` ile yerel skill'ler
agentmemory, Claude-Code-tarzı `<dir>/SKILL.md` formatında 17 skill sunar: 9 çağrılabilir aksiyon skill'i (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) ve ajanın gerektiğinde yüklediği 8 referans skill (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Referans skill'ler, kaynaktan üretilen veri tablolarını taşır, bu yüzden asla sapmazlar. vercel-labs'ın [`skills`](https://npmjs.com/package/skills) CLI'si, bunları çağıran ajanın yerel skill dizinine 50+ ajanda (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf ve daha fazlası) otomatik olarak kurar:
@@ -844,7 +636,7 @@ agentmemory girdisi, `mcpServers` şeklini kullanan her host'ta (Cursor, Claude
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks`, paylaşılan customization dizinine MCP ve yakalama hook'larını kurar. Bkz. [Antigravity kurulumu ve sınırları](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks`, güncel IDE sürümleriyle aynı MCP ve hook yapılandırmasını kullanır. Mevcut kurulumlar `--force` ile yenilenmelidir; bkz. [yükseltme notları](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro`, kullanıcı seviyesindeki yapılandırmayı yazar. Workspace geçersiz kılmaları, kodunuzun yanındaki `.kiro/settings/mcp.json` içine gider. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp`, standart `mcpServers` bloğunu yazar. Warp, ayrıca `.claude/skills/`'ten skill'leri otomatik keşfeder; Claude Code eklentisi kurulduğunda 8 agentmemory skill'i (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`), Warp'ın slash-komut paletinde yerli olarak görünür. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp`, standart `mcpServers` bloğunu yazar. Warp, ayrıca `.claude/skills/`'ten skill'leri otomatik keşfeder; Claude Code eklentisi kurulduğunda 9 agentmemory skill'i (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`), Warp'ın slash-komut paletinde yerli olarak görünür. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline`, standart `mcpServers` bloğunu yazar. VS Code eklenti kullanıcıları: aynı bloğu Cline Settings → MCP Servers → Edit JSON üzerinden yapıştırın. |
| **Continue.dev** | `~/.continue/config.yaml` (tercih edilen) veya `config.json` (eski/legacy) | `agentmemory connect continue`, yapılandırma yoksa `config.yaml` oluşturur veya mevcut `config.json` dosyasını değiştirir. Mevcut YAML için yorumları ve çapaları koruyarak `mcpServers:` altına eklenecek bloğu gösterir. Continue nesne yerine dizi kullanır. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed`, `context_servers` altına yazar (Zed'in key'i, `mcpServers` DEĞİL). Uzak MCP sunucuları, bunun yerine `{"url": "..."}` üzerinden bağlanabilir. |
@@ -926,7 +718,6 @@ Her AI kodlama ajanı, yerleşik bellekle gelir: Claude Code'da `MEMORY.md`, Cur
|---|---|---|
| Bakım | Elle düzenlenen talimatlar | İndekslenmiş oturum geçmişi ve anılar |
| Arama | Her şeyi bağlama yükler | BM25 + vektör + graf (yalnızca top-K) |
-| Token maliyeti | 240 gözlemde 22K+ | ~1,900 token (92% daha az) |
| Ajanlar arası | Ajan başına dosyalar | MCP + REST (herhangi bir ajan) |
| Koordinasyon | Yok | Lease'ler, signal'ler, action'lar, routine'ler |
| Gözlemlenebilirlik | Dosyaları manuel okuma | :3113'te gerçek zamanlı görüntüleyici |
@@ -939,7 +730,7 @@ Her AI kodlama ajanı, yerleşik bellekle gelir: Claude Code'da `MEMORY.md`, Cur
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1024,7 +815,7 @@ Reciprocal Rank Fusion (RRF, k=60) ile füzyonlanır ve oturum bazında çeşitl
Bir vektör indeksi doldurulduğunda, `mem::search` (`memory_recall`'un arkasında) hybrid BM25 + vektör ranker'ı kullanır. Embedding olmadan BM25 kullanır. `smart-search`, anahtarsız modda dahi graf verisi mevcut olduğunda ek olarak yapısal graf eşleşmelerini füzyonlayabilir. Lesson recall'u, her sorguda tüm corpus'u taramak yerine özel bir bellek içi (in-memory) BM25 indeksinde çalışır. Supersede edilmiş bellek sürümleri her recall yolundan hariç tutulur; sürüm zinciri geçmişlerini korur.
-Vektörler bir çökme veya force-kill'den sağ çıkar. Vektör indeksi, en fazla her `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`'de (10 dakika) bucket'lar halinde kaydedilir. Bu aralıkta eklenen veya kaldırılan her vektör, state store'daki küçük bir bekleyen (pending) log'a da hemen yazılır ve bir sonraki başlatma, embedding sağlayıcısını çağırmadan bunu replay eder. Her başarılı kayıt, log'u boşaltır. Replay'den sonra hâlâ vektörü olmayan dokümanlar, hiçbiri kalmayana kadar `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) parti halinde arka planda yeniden embed edilir ve durdurulan bir backfill, bir sonraki başlatmada devam eder. `/agentmemory/status` ve görüntüleyici, bekleyen log boyutunu ve backfill durumunu gösterir. Anahtarsız kurulumlar hiçbir şey yazmaz.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25, Yunanca, Kiril, İbranice, Arapça ve aksanlı Latin'i kutudan çıktığı gibi tokenize eder. Çince / Japonca / Korece bellekler için, CJK run'larını kelime seviyesinde token'lara bölmek üzere opsiyonel segmenter'ları kurun (`npm install @node-rs/jieba tiny-segmenter`); bunlar olmadan agentmemory, tüm-run tokenizasyonuna yumuşak geçiş yapar (soft-falls) ve stderr'de tek seferlik bir ipucu yazdırır.
@@ -1040,9 +831,9 @@ Normal npm kurulumu, opsiyonel `@huggingface/transformers` runtime'ını içerir
| Sağlayıcı | Model | Maliyet | Notlar |
|---|---|---|---|
-| **Yerel (önerilen opt-in)** | `all-MiniLM-L6-v2` | Ücretsiz | İlk model indirmesinden sonra cihaz üzerinde, yalnızca BM25'e göre +8pp recall |
+| **Yerel (önerilen opt-in)** | `all-MiniLM-L6-v2` | Ücretsiz | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Ücretsiz katman | 100+ dil, 768/1536/3072 boyut (MRL), 2048-token girdi. `text-embedding-004`'ün yerini alır ([kullanımdan kaldırıldı, 14 Oca 2026'da durduruldu](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | En yüksek kalite |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Ücretli | Kod için optimize edilmiş |
| Cohere | `embed-english-v3.0` | Ücretsiz deneme | Genel amaçlı |
| OpenRouter | Herhangi bir model | Değişken | Çoklu-model proxy |
diff --git a/READMEs/README.uk-UA.md b/READMEs/README.uk-UA.md
index 1f142d48..2c67111f 100644
--- a/READMEs/README.uk-UA.md
+++ b/READMEs/README.uk-UA.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% точність пошуку R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% менше токенів" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 інструменти MCP" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 автоматичних хуків" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 зовнішніх БД" height="38" /></picture>
@@ -333,223 +331,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Бенчмарки" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Точність пригадування
-
-**coding-agent-life-v1** (власний корпус, відтворюваний у sandbox)
-
-| Адаптер | P@5 | R@5 | Частка влучань у топ-5 | латентність p50 |
-|---|---|---|---|---|
-| **agentmemory (гібридний)** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| базовий варіант grep | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-100% влучань у топ-5 на рівні **математичної межі P@5** для цього корпусу (0.240, див. scorecard). Гібридний режим знаходить усі золоті сесії; grep пропускає 1 з 2 золотих результатів у мультисесійному темпоральному запиті. Прирощення — це **пригадування + темпоральність**, а не агрегована точність. Цей бенчмарк невеликий і має мало золотих прикладів; більший LongMemEval-S нижче краще розрізняє результати. Повна розбивка за типами + примітка про виправлення: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 запитань)
+Актуальні результати пошуку, визначення метрик і кроки відтворення:
-| Система | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| резервний варіант лише на BM25 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Економія токенів
-
-| Підхід | Токенів/рік | Вартість/рік |
-|---|---|---|
-| Вставка повного контексту | 19.5M+ | Неможливо (перевищує вікно контексту) |
-| Підсумовування через LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + локальні ембединги | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Модель ембедингів: `all-MiniLM-L6-v2` (локальна, безкоштовна, без API-ключа). Повні звіти: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). Порівняння з конкурентами: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) охоплює agentmemory порівняно з mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Відтворити локально:** [`eval/README.md`](../eval/README.md) — придатний до підключення адаптерів harness для LongMemEval `_s` (публічний набір із 500 запитань) + `coding-agent-life-v1` (власний корпус із 15 сесій). Адаптери grep / vector / agentmemory оцінюються поруч, вивід у форматі NDJSON, опубліковані scorecard потрапляють у [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Добре поєднується з [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything) та [Graphify](https://github.com/safishamsi/graphify).** Індексування графа коду, конвеєри збирання для кількох агентів і ширші графи знань за документами / PDF / зображеннями / відео. agentmemory запам'ятовує роботу; ці три проєкти оживляють решту контекстного шару. Рецепти + таблиця маршрутизації питань: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="Порівняння з конкурентами" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Вбудована (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Тип</strong></td>
-<td>Двигун пам'яті + MCP-сервер</td>
-<td>API шару пам'яті</td>
-<td>Повноцінний рантайм агента</td>
-<td>Персональний AI</td>
-<td>API пам'яті + застосунок</td>
-<td>Хаб командної пам'яті (LLM-проксі)</td>
-<td>Векторна пам'ять (OSS)</td>
-<td>Двигун пам'яті (Oracle DB)</td>
-<td>Система пам'яті</td>
-<td>Статичний файл</td>
-</tr>
-<tr>
-<td><strong>Пригадування R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>Н/Д</td>
-<td>Заявлено самостійно</td>
-<td>PersonaMem 76% (заявлено самостійно)</td>
-<td>~96.6% (заявлено самостійно)</td>
-<td>94.4% (заявлено самостійно)</td>
-<td>Н/Д</td>
-<td>Н/Д (grep)</td>
-</tr>
-<tr>
-<td><strong>Автозахоплення</strong></td>
-<td>12 хуків (без ручних дій)</td>
-<td>Ручні виклики <code>add()</code></td>
-<td>Самостійне редагування агентом</td>
-<td>Вручну</td>
-<td>Видобування на стороні API</td>
-<td>Перехоплення через проксі (підміна base-URL)</td>
-<td>Вручну</td>
-<td>Видобування через API</td>
-<td>Вручну</td>
-<td>Ручне редагування</td>
-</tr>
-<tr>
-<td><strong>Пошук</strong></td>
-<td>BM25 + вектор + граф (злиття RRF)</td>
-<td>Вектор + граф</td>
-<td>Вектор (архівний)</td>
-<td>Семантичний</td>
-<td>Вектор + RAG</td>
-<td>4 типи активів (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Лише вектор</td>
-<td>Вектор + семантичний</td>
-<td>Зважений за згасанням</td>
-<td>Завантажує все у контекст</td>
-</tr>
-<tr>
-<td><strong>Мультиагентність</strong></td>
-<td>MCP + REST + оренди + сигнали</td>
-<td>API (без координації)</td>
-<td>Лише в межах рантайму Letta</td>
-<td>Ні</td>
-<td>Ні</td>
-<td>Командні ролі + спільні активи</td>
-<td>Ні</td>
-<td>Лише в межах скоупу</td>
-<td>Спільна для кількох агентів</td>
-<td>Файли для кожного агента</td>
-</tr>
-<tr>
-<td><strong>Прив'язка до фреймворку</strong></td>
-<td>Немає (будь-який MCP-клієнт)</td>
-<td>Немає</td>
-<td>Висока (потрібно використовувати Letta)</td>
-<td>Автономний</td>
-<td>Немає</td>
-<td>Проксі обробляє кожен виклик моделі</td>
-<td>Немає</td>
-<td>Oracle Database</td>
-<td>Немає</td>
-<td>Формат для кожного агента</td>
-</tr>
-<tr>
-<td><strong>Зовнішні залежності</strong></td>
-<td>Окрема служба бази даних не потрібна</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + векторна БД</td>
-<td>Декілька</td>
-<td>Керована хмара</td>
-<td>Docker-стек (Core + Hub + Proxy)</td>
-<td>Векторне сховище</td>
-<td>Oracle AI Database</td>
-<td>Немає</td>
-<td>Немає</td>
-</tr>
-<tr>
-<td><strong>Життєвий цикл пам'яті</strong></td>
-<td>4-рівнева консолідація + згасання + автозабування</td>
-<td>Пасивне видобування</td>
-<td>Керується агентом</td>
-<td>Вручну</td>
-<td>Автозабування</td>
-<td>Ручний перегляд; автоматична маршрутизація в розробці</td>
-<td>Немає</td>
-<td>Не вказано</td>
-<td>Згасання + консолідація</td>
-<td>Ручне очищення</td>
-</tr>
-<tr>
-<td><strong>Ефективність токенів</strong></td>
-<td>~1,900 токенів/сесію ($10/рік)</td>
-<td>Залежить від інтеграції</td>
-<td>Основна пам'ять у контексті</td>
-<td>Залежить</td>
-<td>Хмарне ціноутворення</td>
-<td>Не вказано</td>
-<td>Без бюджету токенів</td>
-<td>На основі LLM (залежить)</td>
-<td>Залежить</td>
-<td>22K+ токенів при 240 спостереженнях</td>
-</tr>
-<tr>
-<td><strong>Переглядач у реальному часі</strong></td>
-<td>Так (порт 3113)</td>
-<td>Хмарна панель</td>
-<td>Хмарна панель</td>
-<td>Веб-інтерфейс</td>
-<td>Хмарна панель</td>
-<td>Веб-інтерфейс Hub</td>
-<td>Ні</td>
-<td>Ні</td>
-<td>Ні</td>
-<td>Ні</td>
-</tr>
-<tr>
-<td><strong>Самостійний хостинг</strong></td>
-<td>Так (за замовчуванням)</td>
-<td>Опційно</td>
-<td>Опційно</td>
-<td>Так</td>
-<td>Ні (лише хмара)</td>
-<td>Так (Docker)</td>
-<td>Так</td>
-<td>Так (Oracle DB)</td>
-<td>Так</td>
-<td>Так</td>
-</tr>
-</table>
-
-<sub>Примітка до бенчмарків: лише R@5 для agentmemory — це наш власний вимірений результат (LongMemEval-S, відтворюваний з <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Показники mem0 і Letta — це їхні опубліковані числа LoCoMo (інший набір даних); показники MemPalace, supermemory, TencentDB (PersonaMem) та oracleagentmemory — це заявлені самими постачальниками твердження, які ми не відтворювали самостійно (запуск oracleagentmemory використовував GPT-5.5 проти Oracle AI Database). Наведено поруч лише для орієнтовного порівняння, а не як пряме протистояння на однакових даних. Кількість зірок приблизна і змінюється з часом.</sub>
-
-**Новіші учасники**, варті уваги, детально порівняні в [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Система | ⭐ | Підхід |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Темпоральний граф знань; найкращі опубліковані результати для темпоральних запитів (LongMemEval 63.8%), але граф будується асинхронно, тому нові факти можуть надходити із затримкою |
-| Cognee | 30K | Перетворення документів на граф знань, лише Python, створено для структурованого видобування сутностей, а не для захоплення сесій |
+Оновлене порівняння продуктів і варіантів розгортання з джерелами:
-AgentMemory поєднує захоплення сесій, пошук в історії проєкту й локальний viewer; робота без ключа доступна з першого запуску.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -786,7 +578,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
Запустіть сервер пам'яті: `npx -y @agentmemory/agentmemory@latest`
-#### Нативні скіли через `npx skills add` (50+ агентів)
+#### Нативні скіли через `npx skills add`
agentmemory постачає 17 скілів у форматі `<dir>/SKILL.md` у стилі Claude Code: 9 викликаних скілів-дій (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) і 8 довідкових скілів, які агент завантажує за потреби (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Довідкові скіли містять таблиці даних, згенеровані з джерела, тому вони ніколи не розходяться з ним. CLI [`skills`](https://npmjs.com/package/skills) від vercel-labs автоматично встановлює їх у нативний каталог скілів агента, що викликає, для понад 50 агентів (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf та інших):
@@ -844,7 +636,7 @@ npx skills add rohitg00/agentmemory -y -a '*' # install to every installed age
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` встановлює MCP і хуки захоплення у спільному каталозі налаштувань. Див. [налаштування та обмеження Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` використовує ту саму конфігурацію MCP і хуків, що й поточні версії IDE. Наявні встановлення слід оновити за допомогою `--force`; див. [примітки щодо оновлення](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` записує конфігурацію рівня користувача. Перевизначення на рівні робочого простору розміщуються в `.kiro/settings/mcp.json` поруч із вашим кодом. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` записує стандартний блок `mcpServers`. Warp також автоматично виявляє скіли з `.claude/skills/`; після встановлення плагіна Claude Code 8 скілів agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) з'являються нативно в палітрі слеш-команд Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` записує стандартний блок `mcpServers`. Warp також автоматично виявляє скіли з `.claude/skills/`; після встановлення плагіна Claude Code 9 скілів agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) з'являються нативно в палітрі слеш-команд Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` записує стандартний блок `mcpServers`. Користувачі розширення VS Code: вставте той самий блок через Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (бажано) або `config.json` (застарілий) | `agentmemory connect continue` створює `config.yaml`, якщо налаштувань немає, або змінює наявний `config.json`. Для наявного YAML показує блок для `mcpServers:`, зберігаючи коментарі та якорі. Continue використовує масив, а не об’єкт. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` записує дані під `context_servers` (ключ Zed, НЕ `mcpServers`). Віддалені MCP-сервери замість цього можна підключити через `{"url": "..."}`. |
@@ -927,7 +719,6 @@ REST використовує `3111`, viewer `3113`. Залиште їх у пр
|---|---|---|
| Обслуговування | Інструкції, редаговані вручну | Індексована історія сесій і спогади |
| Пошук | Завантажує все у контекст | BM25 + вектор + граф (лише топ-K) |
-| Вартість у токенах | 22K+ при 240 спостереженнях | ~1,900 токенів (на 92% менше) |
| Між агентами | Файли для кожного агента | MCP + REST (будь-який агент) |
| Координація | Немає | Оренди, сигнали, дії, рутини |
| Спостережуваність | Читання файлів вручну | Переглядач у реальному часі на :3113 |
@@ -940,7 +731,7 @@ REST використовує `3111`, viewer `3113`. Залиште їх у пр
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1025,7 +816,7 @@ SessionStart hook fires
Коли векторний індекс заповнено, `mem::search` (що стоїть за `memory_recall`) використовує гібридний ранжувальник BM25 + вектор. Без ембедингів він використовує BM25. `smart-search` може додатково об'єднувати структурні збіги графа, якщо дані графа існують, зокрема в режимі без ключа. Пригадування уроків виконується на окремому BM25-індексі в пам'яті, а не шляхом сканування всього корпусу під час кожного запиту. Заміщені версії пам'яті виключаються з усіх шляхів пригадування; ланцюжок версій зберігає їхню історію.
-Вектори переживають збій або примусове завершення. Векторний індекс зберігається пакетами не частіше, ніж кожні `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 хвилин). Кожен вектор, доданий або видалений у проміжку, також відразу записується у невеликий журнал очікування в хранилищі стану, і наступний запуск відтворює його без звернення до постачальника ембедингів. Кожне успішне збереження очищає журнал. Документи, які все ще не мають вектора після відтворення, повторно ембедяться у фоновому режимі партіями по `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500), доки не залишиться жодного, а зупинене доповнення продовжується з наступного запуску. `/agentmemory/status` і переглядач показують розмір журналу очікування та стан доповнення. Встановлення без ключа нічого не записують.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 одразу токенізує грецьку, кирилицю, іврит, арабську та латиницю з діакритикою. Для пам'яті китайською / японською / корейською встановіть опційні сегментатори (`npm install @node-rs/jieba tiny-segmenter`), щоб розбивати послідовності CJK на токени рівня слів; без них agentmemory плавно переходить на токенізацію цілими послідовностями та одноразово виводить підказку в stderr.
@@ -1041,9 +832,9 @@ EMBEDDING_PROVIDER=local
| Постачальник | Модель | Вартість | Примітки |
|---|---|---|---|
-| **Локальний (рекомендоване підключення)** | `all-MiniLM-L6-v2` | Безкоштовно | На пристрої після першого завантаження моделі, +8пп до пригадування порівняно з лише BM25 |
+| **Локальний (рекомендоване підключення)** | `all-MiniLM-L6-v2` | Безкоштовно | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Безкоштовний тариф | 100+ мов, 768/1536/3072 вимірів (MRL), вхід до 2048 токенів. Замінює `text-embedding-004` ([застаріла, вимкнення 14 січня 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Найвища якість |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Платно | Оптимізовано для коду |
| Cohere | `embed-english-v3.0` | Безкоштовна пробна версія | Загального призначення |
| OpenRouter | Будь-яка модель | Залежить | Проксі для декількох моделей |
diff --git a/READMEs/README.ur-PK.md b/READMEs/README.ur-PK.md
index 9ea6efaf..fd79b858 100644
--- a/READMEs/README.ur-PK.md
+++ b/READMEs/README.ur-PK.md
@@ -68,8 +68,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% ریٹریول R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% کم ٹوکنز" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP ٹولز" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 خودکار ہکس" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 خارجی ڈیٹابیسز" height="38" /></picture>
@@ -333,223 +331,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="بینچ مارکس" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### ریٹریول کی درستگی
-
-**coding-agent-life-v1** (اندرونی corpus، sandbox میں دوبارہ قابلِ تولید)
-
-| Adapter | P@5 | R@5 | ٹاپ-5 ہِٹ ریٹ | p50 لیٹینسی |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep بیس لائن | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-اس corpus کے لیے **P@5 ریاضیاتی حد (ceiling)** پر 100% ٹاپ-5 ہِٹ ریٹ (0.240، scorecard دیکھیں)۔ Hybrid ہر gold سیشن کو ریٹریو کرتا ہے؛ grep multi-session temporal کوئری پر 2 میں سے 1 gold مِس کرتا ہے۔ فائدہ **recall + temporal** کا ہے، مجموعی precision کا نہیں۔ یہ بینچ مارک چھوٹا اور gold-sparse ہے؛ نیچے دیا گیا بڑا LongMemEval-S بہتر فرق دکھاتا ہے۔ مکمل فی-ٹائپ بریک ڈاؤن + تصحیحی نوٹ: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)۔
-
-**LongMemEval-S** (ICLR 2025، 500 سوالات)
+موجودہ بازیافت کے نتائج، پیمانوں کی تعریفیں اور دوبارہ جانچ کے مراحل:
-| سسٹم | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only فال بیک | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### ٹوکن کی بچت
-
-| طریقہ کار | ٹوکنز/سال | لاگت/سال |
-|---|---|---|
-| مکمل context پیسٹ کرنا | 19.5M+ | ناممکن (window سے تجاوز) |
-| LLM-summarized | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + مقامی embeddings | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding ماڈل: `all-MiniLM-L6-v2` (مقامی، مفت، کوئی API key نہیں)۔ مکمل رپورٹس: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md)، [`benchmark/QUALITY.md`](../benchmark/QUALITY.md)، [`benchmark/SCALE.md`](../benchmark/SCALE.md)۔ حریف موازنہ: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) جو agentmemory کا mem0، Letta، Khoj، supermemory، TencentDB Agent Memory، MemPalace، Zep/Graphiti، Cognee، Hippo سے موازنہ کرتا ہے۔
-
-**مقامی طور پر دہرائیں:** [`eval/README.md`](../eval/README.md)، LongMemEval `_s` (عوامی 500-سوالات) + `coding-agent-life-v1` (اندرونی 15-سیشن corpus) کے لیے ایک adapter-pluggable ہارنس۔ Grep / vector / agentmemory adapters ساتھ ساتھ اسکور کرتے ہیں، NDJSON آؤٹ پٹ، شائع شدہ scorecards [`docs/benchmarks/`](../docs/benchmarks/) میں آتے ہیں۔
-
-**[codegraph](https://github.com/colbymchenry/codegraph)، [Understand Anything](https://github.com/Lum1104/Understand-Anything)، اور [Graphify](https://github.com/safishamsi/graphify) کے ساتھ جوڑی بناتا ہے۔** Code-graph انڈیکسنگ، multi-agent build pipelines، اور docs / PDFs / images / videos میں وسیع نالج گرافس۔ agentmemory کام کو یاد رکھتا ہے؛ یہ تین پراجیکٹس باقی context layer کو روشن کرتے ہیں۔ Recipes + سوال-روٹنگ ٹیبل: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md)۔
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="حریفوں کے مقابلے میں" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>بلٹ-اِن (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>قسم</strong></td>
-<td>میموری انجن + MCP سرور</td>
-<td>میموری لیئر API</td>
-<td>مکمل ایجنٹ رن ٹائم</td>
-<td>ذاتی AI</td>
-<td>میموری API + ایپ</td>
-<td>ٹیم میموری ہب (LLM پراکسی)</td>
-<td>ویکٹر میموری (OSS)</td>
-<td>میموری انجن (Oracle DB)</td>
-<td>میموری سسٹم</td>
-<td>سٹیٹک فائل</td>
-</tr>
-<tr>
-<td><strong>ریٹریول R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>خود رپورٹ شدہ</td>
-<td>PersonaMem 76% (خود رپورٹ شدہ)</td>
-<td>~96.6% (خود رپورٹ شدہ)</td>
-<td>94.4% (خود رپورٹ شدہ)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>خودکار کیپچر</strong></td>
-<td>12 ہکس (کوئی دستی کام نہیں)</td>
-<td>دستی `add()` کالز</td>
-<td>ایجنٹ خود ترمیم کرتا ہے</td>
-<td>دستی</td>
-<td>API کی جانب سے استخراج</td>
-<td>پراکسی انٹرسیپشن (base-URL سویپ)</td>
-<td>دستی</td>
-<td>API استخراج</td>
-<td>دستی</td>
-<td>دستی ترمیم</td>
-</tr>
-<tr>
-<td><strong>سرچ</strong></td>
-<td>BM25 + ویکٹر + گراف (RRF فیوژن)</td>
-<td>ویکٹر + گراف</td>
-<td>ویکٹر (آرکائیول)</td>
-<td>سیمینٹک</td>
-<td>ویکٹر + RAG</td>
-<td>4 asset اقسام (Chat / Skill / Wiki / CodeGraph)</td>
-<td>صرف ویکٹر</td>
-<td>ویکٹر + سیمینٹک</td>
-<td>زوال پر مبنی وزن</td>
-<td>سب کچھ context میں لوڈ کرتا ہے</td>
-</tr>
-<tr>
-<td><strong>ملٹی-ایجنٹ</strong></td>
-<td>MCP + REST + لیز + سگنلز</td>
-<td>API (کوئی ربط نہیں)</td>
-<td>صرف Letta رن ٹائم کے اندر</td>
-<td>نہیں</td>
-<td>نہیں</td>
-<td>ٹیم کردار + مشترک asset</td>
-<td>نہیں</td>
-<td>صرف محدود دائرہ کار میں</td>
-<td>ملٹی-ایجنٹ مشترک</td>
-<td>فی-ایجنٹ فائلیں</td>
-</tr>
-<tr>
-<td><strong>فریم ورک لاک-اِن</strong></td>
-<td>کوئی نہیں (کوئی بھی MCP کلائنٹ)</td>
-<td>کوئی نہیں</td>
-<td>زیادہ (لازمی طور پر Letta استعمال کرنا ہوگا)</td>
-<td>اسٹینڈ الون</td>
-<td>کوئی نہیں</td>
-<td>ہر model کال کے سامنے پراکسی ہوتی ہے</td>
-<td>کوئی نہیں</td>
-<td>Oracle Database</td>
-<td>کوئی نہیں</td>
-<td>فی-ایجنٹ فارمیٹ</td>
-</tr>
-<tr>
-<td><strong>خارجی انحصار</strong></td>
-<td>الگ ڈیٹابیس سروس درکار نہیں</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + ویکٹر DB</td>
-<td>متعدد</td>
-<td>مینیجڈ کلاؤڈ</td>
-<td>Docker اسٹیک (Core + Hub + Proxy)</td>
-<td>ویکٹر اسٹور</td>
-<td>Oracle AI Database</td>
-<td>کوئی نہیں</td>
-<td>کوئی نہیں</td>
-</tr>
-<tr>
-<td><strong>میموری لائف سائیکل</strong></td>
-<td>4-درجاتی استحکام + زوال + خودکار فراموشی</td>
-<td>غیر فعال استخراج</td>
-<td>ایجنٹ کے زیرِ انتظام</td>
-<td>دستی</td>
-<td>خودکار فراموشی</td>
-<td>دستی جائزہ؛ auto-routing زیرِ تکمیل</td>
-<td>کوئی نہیں</td>
-<td>واضح نہیں</td>
-<td>زوال + استحکام</td>
-<td>دستی تراش</td>
-</tr>
-<tr>
-<td><strong>ٹوکن کارکردگی</strong></td>
-<td>~1,900 ٹوکنز/سیشن ($10/سال)</td>
-<td>integration کے لحاظ سے مختلف</td>
-<td>بنیادی میموری context میں</td>
-<td>مختلف ہوتا ہے</td>
-<td>کلاؤڈ قیمتیں</td>
-<td>واضح نہیں</td>
-<td>کوئی token بجٹ نہیں</td>
-<td>LLM پر مبنی (مختلف ہوتا ہے)</td>
-<td>مختلف ہوتا ہے</td>
-<td>240 obs پر 22K+ ٹوکنز</td>
-</tr>
-<tr>
-<td><strong>ریئل-ٹائم ویور</strong></td>
-<td>ہاں (پورٹ 3113)</td>
-<td>کلاؤڈ ڈیش بورڈ</td>
-<td>کلاؤڈ ڈیش بورڈ</td>
-<td>ویب UI</td>
-<td>کلاؤڈ ڈیش بورڈ</td>
-<td>ہب ویب UI</td>
-<td>نہیں</td>
-<td>نہیں</td>
-<td>نہیں</td>
-<td>نہیں</td>
-</tr>
-<tr>
-<td><strong>سیلف-ہوسٹڈ</strong></td>
-<td>ہاں (ڈیفالٹ)</td>
-<td>اختیاری</td>
-<td>اختیاری</td>
-<td>ہاں</td>
-<td>نہیں (صرف کلاؤڈ)</td>
-<td>ہاں (Docker)</td>
-<td>ہاں</td>
-<td>ہاں (Oracle DB)</td>
-<td>ہاں</td>
-<td>ہاں</td>
-</tr>
-</table>
-
-<sub>بینچ مارک نوٹ: صرف agentmemory کا R@5 ہمارا اپنا پیمائش شدہ نتیجہ ہے (LongMemEval-S، <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> سے دوبارہ قابلِ تولید)۔ mem0 اور Letta کے اعداد و شمار ان کے شائع شدہ LoCoMo نمبرز ہیں (ایک مختلف dataset)؛ MemPalace، supermemory، TencentDB (PersonaMem)، اور oracleagentmemory کے اعداد vendor کے خود رپورٹ شدہ دعوے ہیں جنہیں ہم نے آزادانہ طور پر دوبارہ تولید نہیں کیا (oracleagentmemory کا رن Oracle AI Database کے خلاف GPT-5.5 استعمال کرتا تھا)۔ محض اندازاً موازنے کے لیے ساتھ ساتھ دکھایا گیا ہے، ایک جیسے ڈیٹا پر براہِ راست مقابلہ نہیں۔ Star کی تعداد تخمینی ہے اور وقت کے ساتھ بدلتی رہتی ہے۔</sub>
-
-**نئے داخل ہونے والے** جن کے بارے میں جاننا مفید ہے، تفصیلی موازنہ [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) میں:
-
-| سسٹم | ⭐ | زاویہ |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Temporal نالج گراف؛ شائع شدہ temporal-query نتائج میں سب سے مضبوط (LongMemEval 63.8%)، لیکن گراف asynchronously بنتا ہے لہٰذا تازہ facts تاخیر سے آ سکتے ہیں |
-| Cognee | 30K | Document-to-knowledge-graph ingestion، صرف Python، structured entity extraction کے لیے بنایا گیا ہے، سیشن کیپچر کے لیے نہیں |
+مصنوعات اور تنصیب کے طریقوں کا تازہ موازنہ، حوالوں کے ساتھ:
-AgentMemory سیشن capture، قابل تلاش منصوبے کی تاریخ اور مقامی ویوئر اکٹھے فراہم کرتا ہے، اور پہلی بار سے بغیر کلید کام کر سکتا ہے۔
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -782,7 +574,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
میموری سرور شروع کریں: `npx -y @agentmemory/agentmemory@latest`
-#### `npx skills add` کے ذریعے نیٹیو اسکلز (50+ ایجنٹس)
+#### `npx skills add` کے ذریعے نیٹیو اسکلز
agentmemory 17 اسکلز کو Claude-Code طرز کے `<dir>/SKILL.md` فارمیٹ میں شپ کرتا ہے: 9 invocable ایکشن اسکلز (`remember`، `recall`، `recap`، `handoff`، `forget`، `lesson`، `commit-context`، `commit-history`، `session-history`) اور 8 reference اسکلز جنہیں ایجنٹ ضرورت پڑنے پر لوڈ کرتا ہے (`memory-discipline`، `agentmemory-mcp-tools`، `agentmemory-rest-api`، `agentmemory-config`، `agentmemory-agents`، `agentmemory-hooks`، `agentmemory-architecture`، `write-agentmemory-skill`)۔ reference اسکلز سورس سے تیار کردہ data tables رکھتی ہیں، لہٰذا یہ کبھی drift نہیں کرتیں۔ vercel-labs کا [`skills`](https://npmjs.com/package/skills) CLI انہیں calling ایجنٹ کی نیٹیو اسکل ڈائرکٹری میں 50+ ایجنٹس (Claude Code، Cursor، Cline، Continue، Droid، Warp، Codex، Antigravity، Kiro، OpenCode، Goose، Roo، Trae، Windsurf، اور مزید) میں خودکار طور پر انسٹال کرتا ہے:
@@ -839,7 +631,7 @@ agentmemory کا entry ہر اس host پر **ایک ہی MCP سرور بلاک**
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` مشترکہ customization ڈائرکٹری میں MCP اور capture hooks انسٹال کرتا ہے۔ دیکھیں [Antigravity سیٹ اپ اور حدود](../docs/plugins/antigravity.md)۔ |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` موجودہ IDE ورژنز جیسی ہی MCP اور hook configuration استعمال کرتا ہے۔ موجودہ انسٹالیشنز کو `--force` کے ساتھ ریفریش کرنا چاہیے؛ دیکھیں [اپ گریڈ نوٹس](../docs/plugins/antigravity.md)۔ |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` user-level کنفیگ لکھتا ہے۔ Workspace overrides آپ کے کوڈ کے ساتھ `.kiro/settings/mcp.json` میں جاتے ہیں۔ |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` معیاری `mcpServers` بلاک لکھتا ہے۔ Warp `.claude/skills/` سے اسکلز کو بھی خودکار طور پر دریافت کرتا ہے؛ ایک بار Claude Code plugin انسٹال ہونے پر 8 agentmemory اسکلز (`remember`، `recall`، `recap`، `handoff`، `forget`، `commit-context`، `commit-history`، `session-history`) Warp کے slash-command palette میں نیٹیو طور پر ظاہر ہوتی ہیں۔ |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` معیاری `mcpServers` بلاک لکھتا ہے۔ Warp `.claude/skills/` سے اسکلز کو بھی خودکار طور پر دریافت کرتا ہے؛ ایک بار Claude Code plugin انسٹال ہونے پر 9 agentmemory اسکلز (`remember`، `recall`، `recap`، `handoff`، `forget`، `lesson`، `commit-context`، `commit-history`، `session-history`) Warp کے slash-command palette میں نیٹیو طور پر ظاہر ہوتی ہیں۔ |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` معیاری `mcpServers` بلاک لکھتا ہے۔ VS Code extension صارفین: وہی بلاک Cline Settings → MCP Servers → Edit JSON کے ذریعے پیسٹ کریں۔ |
| **Continue.dev** | `~/.continue/config.yaml` (ترجیحی) یا `config.json` (legacy) | `agentmemory connect continue` ترتیب نہ ہونے پر `config.yaml` بناتا ہے یا موجودہ `config.json` اپ ڈیٹ کرتا ہے۔ موجودہ YAML کے comments اور anchors محفوظ رکھنے کے لیے اسے تبدیل نہیں کرتا، بلکہ `mcpServers:` کے نیچے چسپاں کرنے کا بلاک دکھاتا ہے۔ Continue کا `mcpServers` ایک array ہے۔ |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` `context_servers` کے تحت لکھتا ہے (Zed کی اپنی key، `mcpServers` نہیں)۔ ریموٹ MCP سرورز کو بجائے `{"url": "..."}` کے ذریعے جوڑا جا سکتا ہے۔ |
@@ -930,7 +722,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| دیکھ بھال | ہاتھ سے ترتیب دی گئی ہدایات | انڈیکس شدہ سیشن کی تاریخ اور میموری |
| سرچ | سب کچھ context میں لوڈ کرتا ہے | BM25 + ویکٹر + گراف (صرف top-K) |
-| ٹوکن لاگت | 240 observations پر 22K+ | ~1,900 ٹوکنز (92% کم) |
| کراس-ایجنٹ | فی-ایجنٹ فائلیں | MCP + REST (کوئی بھی ایجنٹ) |
| کوآرڈینیشن | کوئی نہیں | Leases، سگنلز، actions، routines |
| Observability | دستی طور پر فائلیں پڑھنا | :3113 پر ریئل-ٹائم ویور |
@@ -943,7 +734,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1028,7 +819,7 @@ Reciprocal Rank Fusion (RRF، k=60) کے ساتھ فیوز کیا گیا اور
جب ویکٹر index پاپولیٹڈ ہو، `mem::search` (`memory_recall` کے پیچھے) hybrid BM25 + vector ranker استعمال کرتا ہے۔ embeddings کے بغیر یہ BM25 استعمال کرتا ہے۔ `smart-search` گراف ڈیٹا موجود ہونے پر اضافی طور پر structural graph matches بھی فیوز کر سکتا ہے۔ Lesson recall ہر کوئری پر پورے corpus کو اسکین کرنے کے بجائے ایک وقف شدہ in-memory BM25 index پر چلتا ہے۔ سپرسیڈ شدہ میموری ورژنز ہر recall راستے سے خارج ہوتے ہیں؛ version chain ان کی ہسٹری رکھتی ہے۔
-ویکٹرز کریش یا force-kill سے بچ جاتے ہیں۔ ویکٹر index کم از کم ہر `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 منٹ) بعد buckets میں محفوظ ہوتا ہے۔ اس دوران شامل یا ہٹایا گیا ہر ویکٹر state store میں ایک چھوٹے pending log میں فوراً لکھ دیا جاتا ہے، اور اگلا اسٹارٹ اسے embedding provider کو کال کیے بغیر replay کرتا ہے۔ ہر کامیاب save لاگ کو خالی کر دیتا ہے۔ وہ documents جن کا replay کے بعد بھی کوئی ویکٹر نہیں ہوتا انہیں پسِ منظر میں `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) کے batches میں دوبارہ embed کیا جاتا ہے جب تک کوئی باقی نہ رہے، اور رکا ہوا backfill اگلے اسٹارٹ پر جاری رہتا ہے۔ `/agentmemory/status` اور ویور pending log کا سائز اور backfill اسٹیٹس دکھاتے ہیں۔ Keyless انسٹالز کچھ نہیں لکھتے۔
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 بذاتِ خود یونانی، سیریلک، عبرانی، عربی، اور accented Latin کو ٹوکنائز کرتا ہے۔ چینی / جاپانی / کوریائی میموریز کے لیے، CJK runs کو word-level tokens میں تقسیم کرنے کے لیے اختیاری segmenters (`npm install @node-rs/jieba tiny-segmenter`) انسٹال کریں؛ ان کے بغیر، agentmemory نرمی سے whole-run tokenization پر fall ہو جاتا ہے اور stderr پر ایک بار اشارہ پرنٹ کرتا ہے۔
@@ -1044,9 +835,9 @@ EMBEDDING_PROVIDER=local
| Provider | ماڈل | لاگت | نوٹس |
|---|---|---|---|
-| **Local (تجویز کردہ opt-in)** | `all-MiniLM-L6-v2` | مفت | پہلے ماڈل ڈاؤن لوڈ کے بعد آن-ڈیوائس، صرف-BM25 کے مقابلے میں +8pp recall |
+| **Local (تجویز کردہ opt-in)** | `all-MiniLM-L6-v2` | مفت | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | مفت ٹیئر | 100+ زبانیں، 768/1536/3072 dims (MRL)، 2048-ٹوکن input۔ `text-embedding-004` کی جگہ لیتا ہے ([deprecated، shutdown Jan 14, 2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | بہترین کوالٹی |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | ادا شدہ (Paid) | کوڈ کے لیے optimized |
| Cohere | `embed-english-v3.0` | مفت trial | عمومی مقصد |
| OpenRouter | کوئی بھی ماڈل | مختلف ہوتا ہے | Multi-model پراکسی |
diff --git a/READMEs/README.vi-VN.md b/READMEs/README.vi-VN.md
index 06aea9a2..6a80a5fe 100644
--- a/READMEs/README.vi-VN.md
+++ b/READMEs/README.vi-VN.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% retrieval R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="92% fewer tokens" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 MCP tools" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 auto hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 external DBs" height="38" /></picture>
@@ -331,223 +329,17 @@ Ghi chú phát hành mới nhất: [CHANGELOG.md](../CHANGELOG.md).
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmark" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### Độ chính xác Retrieval
-
-**coding-agent-life-v1** (corpus nội bộ, có thể tái lập trong sandbox)
-
-| Adapter | P@5 | R@5 | Tỷ lệ hit top-5 | Độ trễ p50 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep baseline | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-Tỷ lệ hit top-5 đạt 100% tại **ngưỡng toán học P@5** cho corpus này (0.240, xem scorecard). Hybrid lấy được mọi gold session; grep bỏ lỡ 1 trong 2 gold ở câu truy vấn temporal đa session. Mức cải thiện nằm ở **recall + temporal**, không phải precision tổng hợp. Benchmark này nhỏ và thưa gold; LongMemEval-S lớn hơn ở dưới phân biệt tốt hơn. Phân tích chi tiết theo từng loại + ghi chú đính chính đầy đủ: [`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md).
-
-**LongMemEval-S** (ICLR 2025, 500 câu hỏi)
+Kết quả truy xuất hiện tại, định nghĩa chỉ số và các bước tái lập:
-| Hệ thống | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| BM25-only fallback | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Tiết kiệm Token
-
-| Cách tiếp cận | Token/năm | Chi phí/năm |
-|---|---|---|
-| Dán toàn bộ context | 19.5M+ | Không thể (vượt window) |
-| Tóm tắt bằng LLM | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + embedding cục bộ | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> Embedding model: `all-MiniLM-L6-v2` (cục bộ, miễn phí, không cần API key). Báo cáo đầy đủ: [`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md), [`benchmark/QUALITY.md`](../benchmark/QUALITY.md), [`benchmark/SCALE.md`](../benchmark/SCALE.md). So sánh đối thủ: [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md) bao gồm agentmemory so với mem0, Letta, Khoj, supermemory, TencentDB Agent Memory, MemPalace, Zep/Graphiti, Cognee, Hippo.
-
-**Tái lập cục bộ:** [`eval/README.md`](../eval/README.md), một harness có thể cắm (pluggable) adapter cho LongMemEval `_s` (500 câu hỏi công khai) + `coding-agent-life-v1` (corpus nội bộ 15 session). Các adapter grep / vector / agentmemory được chấm điểm song song, xuất NDJSON, các scorecard đã công bố nằm tại [`docs/benchmarks/`](../docs/benchmarks/).
-
-**Kết hợp tốt với [codegraph](https://github.com/colbymchenry/codegraph), [Understand Anything](https://github.com/Lum1104/Understand-Anything), và [Graphify](https://github.com/safishamsi/graphify).** Index hóa code-graph, pipeline build đa agent, và các knowledge graph rộng hơn trên docs / PDF / ảnh / video. agentmemory ghi nhớ phần công việc; ba dự án này làm sáng rõ phần còn lại của context layer. Công thức + bảng định tuyến câu hỏi: [`docs/recipes/pairings.md`](../docs/recipes/pairings.md).
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="So sánh đối thủ" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>Tích hợp sẵn (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>Loại</strong></td>
-<td>Engine bộ nhớ + MCP server</td>
-<td>API lớp bộ nhớ</td>
-<td>Runtime agent đầy đủ</td>
-<td>AI cá nhân</td>
-<td>API bộ nhớ + app</td>
-<td>Hub bộ nhớ nhóm (LLM proxy)</td>
-<td>Bộ nhớ vector (OSS)</td>
-<td>Engine bộ nhớ (Oracle DB)</td>
-<td>Hệ thống bộ nhớ</td>
-<td>File tĩnh</td>
-</tr>
-<tr>
-<td><strong>Retrieval R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>Tự công bố</td>
-<td>PersonaMem 76% (tự công bố)</td>
-<td>~96.6% (tự công bố)</td>
-<td>94.4% (tự công bố)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>Auto-capture</strong></td>
-<td>12 hooks (không cần thao tác thủ công)</td>
-<td>Gọi <code>add()</code> thủ công</td>
-<td>Agent tự sửa</td>
-<td>Thủ công</td>
-<td>Trích xuất phía API</td>
-<td>Chặn qua proxy (đổi base-URL)</td>
-<td>Thủ công</td>
-<td>Trích xuất qua API</td>
-<td>Thủ công</td>
-<td>Chỉnh sửa thủ công</td>
-</tr>
-<tr>
-<td><strong>Tìm kiếm</strong></td>
-<td>BM25 + Vector + Graph (RRF fusion)</td>
-<td>Vector + Graph</td>
-<td>Vector (lưu trữ archival)</td>
-<td>Ngữ nghĩa (semantic)</td>
-<td>Vector + RAG</td>
-<td>4 loại asset (Chat / Skill / Wiki / CodeGraph)</td>
-<td>Chỉ Vector</td>
-<td>Vector + ngữ nghĩa</td>
-<td>Có trọng số suy giảm (decay)</td>
-<td>Nạp tất cả vào context</td>
-</tr>
-<tr>
-<td><strong>Đa agent</strong></td>
-<td>MCP + REST + lease + signal</td>
-<td>API (không điều phối)</td>
-<td>Chỉ trong runtime Letta</td>
-<td>Không</td>
-<td>Không</td>
-<td>Vai trò nhóm + asset chia sẻ</td>
-<td>Không</td>
-<td>Chỉ theo scope</td>
-<td>Chia sẻ đa agent</td>
-<td>File theo từng agent</td>
-</tr>
-<tr>
-<td><strong>Lệ thuộc framework</strong></td>
-<td>Không (bất kỳ MCP client nào)</td>
-<td>Không</td>
-<td>Cao (phải dùng Letta)</td>
-<td>Độc lập (standalone)</td>
-<td>Không</td>
-<td>Proxy chặn trước mọi lệnh gọi model</td>
-<td>Không</td>
-<td>Oracle Database</td>
-<td>Không</td>
-<td>Định dạng theo từng agent</td>
-</tr>
-<tr>
-<td><strong>Dependency bên ngoài</strong></td>
-<td>Không cần dịch vụ cơ sở dữ liệu riêng</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + vector DB</td>
-<td>Nhiều</td>
-<td>Cloud được quản lý</td>
-<td>Docker stack (Core + Hub + Proxy)</td>
-<td>Vector store</td>
-<td>Oracle AI Database</td>
-<td>Không</td>
-<td>Không</td>
-</tr>
-<tr>
-<td><strong>Lifecycle bộ nhớ</strong></td>
-<td>Hợp nhất 4 tầng + suy giảm + tự quên</td>
-<td>Trích xuất bị động</td>
-<td>Do agent quản lý</td>
-<td>Thủ công</td>
-<td>Tự quên (auto-forget)</td>
-<td>Review thủ công; auto-routing đang phát triển</td>
-<td>Không</td>
-<td>Không công bố</td>
-<td>Suy giảm + hợp nhất</td>
-<td>Cắt bỏ thủ công</td>
-</tr>
-<tr>
-<td><strong>Hiệu quả token</strong></td>
-<td>~1,900 token/session ($10/năm)</td>
-<td>Thay đổi theo tích hợp</td>
-<td>Bộ nhớ lõi (core memory) nằm trong context</td>
-<td>Thay đổi</td>
-<td>Theo giá cloud</td>
-<td>Không công bố</td>
-<td>Không có ngân sách token</td>
-<td>Dựa trên LLM (thay đổi)</td>
-<td>Thay đổi</td>
-<td>22K+ token ở 240 observation</td>
-</tr>
-<tr>
-<td><strong>Viewer real-time</strong></td>
-<td>Có (port 3113)</td>
-<td>Cloud dashboard</td>
-<td>Cloud dashboard</td>
-<td>Web UI</td>
-<td>Cloud dashboard</td>
-<td>Hub web UI</td>
-<td>Không</td>
-<td>Không</td>
-<td>Không</td>
-<td>Không</td>
-</tr>
-<tr>
-<td><strong>Self-hosted</strong></td>
-<td>Có (mặc định)</td>
-<td>Tùy chọn</td>
-<td>Tùy chọn</td>
-<td>Có</td>
-<td>Không (chỉ cloud)</td>
-<td>Có (Docker)</td>
-<td>Có</td>
-<td>Có (Oracle DB)</td>
-<td>Có</td>
-<td>Có</td>
-</tr>
-</table>
-
-<sub>Ghi chú benchmark: chỉ con số R@5 của agentmemory là kết quả đo đạc của chính chúng tôi (LongMemEval-S, có thể tái lập từ <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a>). Các số liệu của mem0 và Letta là số LoCoMo họ đã công bố (một dataset khác); các số liệu của MemPalace, supermemory, TencentDB (PersonaMem), và oracleagentmemory là các tuyên bố tự công bố từ nhà cung cấp mà chúng tôi chưa tái lập độc lập (lần chạy của oracleagentmemory dùng GPT-5.5 trên một Oracle AI Database). Chỉ hiển thị song song để ước lượng, không phải so sánh trực tiếp (head-to-head) trên cùng dữ liệu. Số lượng star chỉ là gần đúng và thay đổi theo thời gian.</sub>
-
-**Những cái tên mới** đáng để biết, được so sánh chi tiết trong [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| Hệ thống | ⭐ | Hướng tiếp cận |
-|--------|---|-------|
-| Zep / Graphiti | 30K | Knowledge graph theo thời gian (temporal); kết quả truy vấn temporal công bố mạnh nhất (LongMemEval 63.8%), nhưng graph được xây dựng bất đồng bộ nên các fact mới có thể bị trễ |
-| Cognee | 30K | Nạp từ tài liệu sang knowledge graph, chỉ hỗ trợ Python, được xây dựng để trích xuất entity có cấu trúc hơn là để capture session |
+So sánh cập nhật về sản phẩm và phương án triển khai, kèm nguồn:
-AgentMemory kết hợp ghi nhận phiên, lịch sử dự án có thể tìm kiếm và trình xem cục bộ, với chế độ không cần khóa ngay từ lần đầu.
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Hướng dẫn đầy đủ: [`integrations/hermes/`](../integrations/hermes/)
Khởi động memory server: `npx -y @agentmemory/agentmemory@latest`
-#### Skill gốc qua `npx skills add` (50+ agent)
+#### Skill gốc qua `npx skills add`
agentmemory đi kèm 17 skill theo định dạng `<dir>/SKILL.md` kiểu Claude Code: 9 skill hành động có thể gọi (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) và 8 skill tham chiếu mà agent nạp theo nhu cầu (`memory-discipline`, `agentmemory-mcp-tools`, `agentmemory-rest-api`, `agentmemory-config`, `agentmemory-agents`, `agentmemory-hooks`, `agentmemory-architecture`, `write-agentmemory-skill`). Các skill tham chiếu mang theo bảng dữ liệu được sinh ra từ source, nên chúng không bao giờ lệch (drift). CLI [`skills`](https://npmjs.com/package/skills) của vercel-labs tự cài chúng vào thư mục skill gốc của agent gọi nó, trên hơn 50 agent (Claude Code, Cursor, Cline, Continue, Droid, Warp, Codex, Antigravity, Kiro, OpenCode, Goose, Roo, Trae, Windsurf, và nhiều hơn nữa):
@@ -837,7 +629,7 @@ Entry của agentmemory là **cùng một block MCP server** trên mọi host d
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` cài MCP và capture hook vào thư mục customization chung. Xem [thiết lập và giới hạn của Antigravity](../docs/plugins/antigravity.md). |
| **Antigravity CLI** (`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` dùng cùng cấu hình MCP và hook như các phiên bản IDE hiện tại. Các bản cài đặt sẵn có nên refresh bằng `--force`; xem [ghi chú upgrade](../docs/plugins/antigravity.md). |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` viết config ở cấp user. Các override ở cấp workspace nằm trong `.kiro/settings/mcp.json` cạnh code của bạn. |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` viết block `mcpServers` chuẩn. Warp cũng tự phát hiện skill từ `.claude/skills/`; khi plugin Claude Code được cài, 8 skill agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `commit-context`, `commit-history`, `session-history`) sẽ xuất hiện một cách tự nhiên trong bảng slash-command của Warp. |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` viết block `mcpServers` chuẩn. Warp cũng tự phát hiện skill từ `.claude/skills/`; khi plugin Claude Code được cài, 9 skill agentmemory (`remember`, `recall`, `recap`, `handoff`, `forget`, `lesson`, `commit-context`, `commit-history`, `session-history`) sẽ xuất hiện một cách tự nhiên trong bảng slash-command của Warp. |
| **Cline (CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` viết block `mcpServers` chuẩn. Người dùng extension VS Code: dán cùng block này qua Cline Settings → MCP Servers → Edit JSON. |
| **Continue.dev** | `~/.continue/config.yaml` (ưu tiên) hoặc `config.json` (legacy) | `agentmemory connect continue` tạo `config.yaml` nếu chưa có cấu hình hoặc cập nhật `config.json` hiện có. YAML có sẵn được giữ nguyên để bảo toàn chú thích và anchor; lệnh in khối để dán dưới `mcpServers:`. Continue dùng dạng mảng cho `mcpServers`. |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` viết dưới `context_servers` (key riêng của Zed, KHÔNG phải `mcpServers`). Các MCP server remote có thể được kết nối qua `{"url": "..."}` thay thế. |
@@ -933,7 +725,6 @@ Mọi AI coding agent đều đi kèm bộ nhớ tích hợp sẵn: Claude Code
|---|---|---|
| Bảo trì | Chỉ dẫn được quản lý thủ công | Lịch sử phiên và bộ nhớ đã lập chỉ mục |
| Tìm kiếm | Nạp tất cả vào context | BM25 + vector + graph (chỉ top-K) |
-| Chi phí token | 22K+ ở 240 observation | ~1,900 token (giảm 92%) |
| Đa agent | File theo từng agent | MCP + REST (mọi agent) |
| Điều phối | Không có | Lease, signal, action, routine |
| Observability | Đọc file thủ công | Viewer real-time trên :3113 |
@@ -946,7 +737,7 @@ Mọi AI coding agent đều đi kèm bộ nhớ tích hợp sẵn: Claude Code
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1031,7 +822,7 @@ Retrieval ba luồng (triple-stream) kết hợp ba tín hiệu:
Khi vector index đã có dữ liệu, `mem::search` (đứng sau `memory_recall`) dùng ranker hybrid BM25 + vector. Khi không có embedding, nó dùng BM25. `smart-search` còn có thể fuse thêm các kết quả khớp từ graph cấu trúc khi dữ liệu graph tồn tại, kể cả ở chế độ keyless. Recall cho lesson chạy trên một BM25 index riêng trong memory thay vì quét toàn bộ corpus cho mỗi truy vấn. Các phiên bản memory đã bị supersede bị loại khỏi mọi đường recall; chuỗi version vẫn giữ lại lịch sử của chúng.
-Vector sống sót qua một lần crash hoặc force-kill. Vector index được lưu theo bucket, nhiều nhất là mỗi `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS` (10 phút). Mỗi vector được thêm hoặc xóa ở giữa cũng được ghi ngay vào một pending log nhỏ trong state store, và lần khởi động kế tiếp sẽ replay nó mà không cần gọi embedding provider. Mỗi lần save thành công sẽ làm trống log đó. Các document vẫn chưa có vector sau khi replay sẽ được re-embed ở nền theo từng batch `AGENTMEMORY_VECTOR_BACKFILL_MAX` (500) cho tới khi không còn document nào, và một backfill bị dừng sẽ tiếp tục ở lần khởi động kế tiếp. `/agentmemory/status` và viewer hiển thị kích thước pending log và trạng thái backfill. Bản cài keyless không ghi gì cả.
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 tokenize được tiếng Hy Lạp, Cyrillic, Hebrew, Ả Rập, và Latin có dấu ngay từ đầu. Với bộ nhớ tiếng Trung / Nhật / Hàn, hãy cài thêm các segmenter tùy chọn (`npm install @node-rs/jieba tiny-segmenter`) để chia các đoạn CJK thành token ở mức từ; nếu không có chúng, agentmemory sẽ lùi nhẹ về tokenize theo cả đoạn và in ra một gợi ý một lần trên stderr.
@@ -1047,9 +838,9 @@ Lần cài npm thông thường đã bao gồm runtime tùy chọn `@huggingface
| Provider | Model | Chi phí | Ghi chú |
|---|---|---|---|
-| **Local (khuyến nghị opt-in)** | `all-MiniLM-L6-v2` | Miễn phí | Chạy trên máy sau lần tải model đầu tiên, +8pp recall so với chỉ dùng BM25 |
+| **Local (khuyến nghị opt-in)** | `all-MiniLM-L6-v2` | Miễn phí | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | Free tier | 100+ ngôn ngữ, 768/1536/3072 chiều (MRL), input 2048 token. Thay thế `text-embedding-004` ([đã deprecated, shutdown 14/01/2026](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | Chất lượng cao nhất |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | Trả phí | Tối ưu cho code |
| Cohere | `embed-english-v3.0` | Free trial | Dùng chung (general purpose) |
| OpenRouter | Bất kỳ model nào | Thay đổi | Proxy đa model |
diff --git a/READMEs/README.zh-CN.md b/READMEs/README.zh-CN.md
index 0941ab0d..17a2acd7 100644
--- a/READMEs/README.zh-CN.md
+++ b/READMEs/README.zh-CN.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% 检索 R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="token 减少 92%" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 个 MCP 工具" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 个自动 hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 个外部数据库" height="38" /></picture>
@@ -331,223 +329,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="基准测试" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### 检索准确率
-
-**coding-agent-life-v1**(内部语料库,可在沙箱中复现)
-
-| 适配器 | P@5 | R@5 | Top-5 命中率 | p50 延迟 |
-|---|---|---|---|---|
-| **agentmemory 混合** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep 基线 | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-在该语料库上,命中率达到了 **P@5 的数学上限**(0.240,详见 scorecard)下的 100% top-5 命中率。混合检索能取回每一个 gold 会话;grep 在多会话时间性查询上漏掉了 2 个 gold 中的 1 个。这里的提升体现在**召回率 + 时间性**,而不是整体精确率。这个基准规模较小、gold 样本稀疏;下方更大规模的 LongMemEval-S 区分度更好。完整的按类型拆分 + 修正说明见:[`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)。
-
-**LongMemEval-S**(ICLR 2025,500 个问题)
+最新检索结果、指标定义及复现步骤:
-| 系统 | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| 仅 BM25 回退 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token 节省
-
-| 方案 | Tokens/年 | 成本/年 |
-|---|---|---|
-| 粘贴完整上下文 | 19.5M+ | 不可行(超出窗口) |
-| LLM 摘要 | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + 本地嵌入 | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> 嵌入模型:`all-MiniLM-L6-v2`(本地、免费、无需 API key)。完整报告见:[`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md)、[`benchmark/QUALITY.md`](../benchmark/QUALITY.md)、[`benchmark/SCALE.md`](../benchmark/SCALE.md)。竞品对比见 [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md),涵盖 agentmemory 与 mem0、Letta、Khoj、supermemory、TencentDB Agent Memory、MemPalace、Zep/Graphiti、Cognee、Hippo 的对比。
-
-**本地复现:** [`eval/README.md`](../eval/README.md) 是一个可插拔适配器的测试框架,面向 LongMemEval `_s`(公开的 500 题)+ `coding-agent-life-v1`(内部 15 会话语料库)。Grep / 向量 / agentmemory 适配器并排评分,输出 NDJSON,发布的 scorecard 存放在 [`docs/benchmarks/`](../docs/benchmarks/)。
-
-**可与 [codegraph](https://github.com/colbymchenry/codegraph)、[Understand Anything](https://github.com/Lum1104/Understand-Anything) 和 [Graphify](https://github.com/safishamsi/graphify) 搭配使用。** 分别覆盖代码图索引、多代理构建流水线,以及跨文档 / PDF / 图像 / 视频的更广泛知识图谱。agentmemory 记住的是工作过程;这三个项目则点亮了上下文层的其余部分。配方 + 问题路由表见:[`docs/recipes/pairings.md`](../docs/recipes/pairings.md)。
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="对比竞品" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>内建 (CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>类型</strong></td>
-<td>记忆引擎 + MCP 服务器</td>
-<td>记忆层 API</td>
-<td>完整代理运行时</td>
-<td>个人 AI</td>
-<td>记忆 API + 应用</td>
-<td>团队记忆中枢(LLM 代理层)</td>
-<td>向量记忆(开源)</td>
-<td>记忆引擎(Oracle DB)</td>
-<td>记忆系统</td>
-<td>静态文件</td>
-</tr>
-<tr>
-<td><strong>检索 R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>自报</td>
-<td>PersonaMem 76%(自报)</td>
-<td>~96.6%(自报)</td>
-<td>94.4%(自报)</td>
-<td>N/A</td>
-<td>N/A (grep)</td>
-</tr>
-<tr>
-<td><strong>自动捕获</strong></td>
-<td>12 hooks(零人工)</td>
-<td>手动调用 <code>add()</code></td>
-<td>代理自编辑</td>
-<td>手动</td>
-<td>API 侧提取</td>
-<td>代理层拦截(替换 base-URL)</td>
-<td>手动</td>
-<td>API 提取</td>
-<td>手动</td>
-<td>手动编辑</td>
-</tr>
-<tr>
-<td><strong>搜索</strong></td>
-<td>BM25 + 向量 + 图(RRF 融合)</td>
-<td>向量 + 图</td>
-<td>向量(归档)</td>
-<td>语义</td>
-<td>向量 + RAG</td>
-<td>4 种资产类型(Chat / Skill / Wiki / CodeGraph)</td>
-<td>仅向量</td>
-<td>向量 + 语义</td>
-<td>衰减加权</td>
-<td>将所有内容加载到上下文</td>
-</tr>
-<tr>
-<td><strong>多代理</strong></td>
-<td>MCP + REST + 租约 + 信号</td>
-<td>API(无协调)</td>
-<td>仅在 Letta 运行时内部</td>
-<td>无</td>
-<td>无</td>
-<td>团队角色 + 共享资产</td>
-<td>无</td>
-<td>仅作用域</td>
-<td>多代理共享</td>
-<td>每代理一个文件</td>
-</tr>
-<tr>
-<td><strong>框架锁定</strong></td>
-<td>无(任何 MCP 客户端)</td>
-<td>无</td>
-<td>高(必须使用 Letta)</td>
-<td>独立</td>
-<td>无</td>
-<td>代理层截获每次模型调用</td>
-<td>无</td>
-<td>Oracle Database</td>
-<td>无</td>
-<td>每代理格式</td>
-</tr>
-<tr>
-<td><strong>外部依赖</strong></td>
-<td>无需独立数据库服务</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + 向量数据库</td>
-<td>多个</td>
-<td>托管云</td>
-<td>Docker 栈(Core + Hub + Proxy)</td>
-<td>向量存储</td>
-<td>Oracle AI Database</td>
-<td>无</td>
-<td>无</td>
-</tr>
-<tr>
-<td><strong>记忆生命周期</strong></td>
-<td>4 层整合 + 衰减 + 自动遗忘</td>
-<td>被动提取</td>
-<td>代理管理</td>
-<td>手动</td>
-<td>自动遗忘</td>
-<td>人工审核;自动路由开发中</td>
-<td>无</td>
-<td>未说明</td>
-<td>衰减 + 整合</td>
-<td>手动清理</td>
-</tr>
-<tr>
-<td><strong>Token 效率</strong></td>
-<td>~1,900 tokens/会话($10/年)</td>
-<td>依集成方式不同</td>
-<td>核心记忆位于上下文</td>
-<td>不定</td>
-<td>云端定价</td>
-<td>未说明</td>
-<td>无 token 预算</td>
-<td>LLM 支撑(不定)</td>
-<td>不定</td>
-<td>240 条观测达 22K+ tokens</td>
-</tr>
-<tr>
-<td><strong>实时查看器</strong></td>
-<td>是(端口 3113)</td>
-<td>云端仪表板</td>
-<td>云端仪表板</td>
-<td>Web UI</td>
-<td>云端仪表板</td>
-<td>Hub Web UI</td>
-<td>无</td>
-<td>无</td>
-<td>无</td>
-<td>无</td>
-</tr>
-<tr>
-<td><strong>自托管</strong></td>
-<td>是(默认)</td>
-<td>可选</td>
-<td>可选</td>
-<td>是</td>
-<td>否(仅云端)</td>
-<td>是(Docker)</td>
-<td>是</td>
-<td>是(Oracle DB)</td>
-<td>是</td>
-<td>是</td>
-</tr>
-</table>
-
-<sub>基准说明:只有 agentmemory 的 R@5 是我们自己测得的结果(LongMemEval-S,可从 <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> 复现)。mem0 和 Letta 的数字是它们公布的 LoCoMo 结果(不同数据集);MemPalace、supermemory、TencentDB(PersonaMem)和 oracleagentmemory 的数字是厂商自报、我们未独立复现的声明(oracleagentmemory 的测试使用 GPT-5.5 搭配 Oracle AI Database)。并列展示仅供粗略参考,并非同一数据上的正面对比。星标数为近似值且会随时间漂移。</sub>
-
-**值得了解的新入局者**,深入对比见 [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| 系统 | ⭐ | 切入角度 |
-|--------|---|-------|
-| Zep / Graphiti | 30K | 时间性知识图谱;已公布的时间性查询结果最强(LongMemEval 63.8%),但图谱是异步构建的,新事实可能滞后 |
-| Cognee | 30K | 文档到知识图谱的摄取,仅 Python,为结构化实体抽取而建,而非会话捕获 |
+产品与部署方式的最新对比及来源:
-AgentMemory 结合会话捕获、可搜索的项目历史和本地查看器,首次使用即可选择无密钥模式。
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -780,7 +572,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
启动记忆服务器:`npx -y @agentmemory/agentmemory@latest`
-#### 通过 `npx skills add` 安装原生 skills(50+ 代理)
+#### 通过 `npx skills add` 安装原生 skills
agentmemory 以 Claude-Code 风格的 `<dir>/SKILL.md` 格式提供 17 个 skills:9 个可调用的操作型 skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)以及 8 个代理按需加载的参考 skills(`memory-discipline`、`agentmemory-mcp-tools`、`agentmemory-rest-api`、`agentmemory-config`、`agentmemory-agents`、`agentmemory-hooks`、`agentmemory-architecture`、`write-agentmemory-skill`)。参考 skills 携带的数据表是从源码生成的,因此不会出现偏差。vercel-labs 开发的 [`skills`](https://npmjs.com/package/skills) CLI 会自动把它们安装到调用方代理的原生 skill 目录,覆盖 50+ 个代理(Claude Code、Cursor、Cline、Continue、Droid、Warp、Codex、Antigravity、Kiro、OpenCode、Goose、Roo、Trae、Windsurf 等):
@@ -837,7 +629,7 @@ Hermes、OpenClaw、Pi 和 Oh My Pi 的连接器默认安装原生集成和 skil
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` 会在共享的自定义目录中安装 MCP 和捕获 hooks。参见[Antigravity 设置与限制](../docs/plugins/antigravity.md)。 |
| **Antigravity CLI**(`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` 使用与当前 IDE 版本相同的 MCP 和 hook 配置。现有安装应使用 `--force` 刷新;参见[升级说明](../docs/plugins/antigravity.md)。 |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` 写入用户级配置。工作区级覆盖放在代码旁边的 `.kiro/settings/mcp.json` 中。 |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` 写入标准的 `mcpServers` 代码块。Warp 还会从 `.claude/skills/` 自动发现 skills;一旦安装了 Claude Code 插件,8 个 agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`commit-context`、`commit-history`、`session-history`)就会原生出现在 Warp 的斜杠命令面板中。 |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` 写入标准的 `mcpServers` 代码块。Warp 还会从 `.claude/skills/` 自动发现 skills;一旦安装了 Claude Code 插件,9 个 agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)就会原生出现在 Warp 的斜杠命令面板中。 |
| **Cline(CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` 写入标准的 `mcpServers` 代码块。VS Code 插件用户:通过 Cline Settings → MCP Servers → Edit JSON 粘贴同样的代码块。 |
| **Continue.dev** | `~/.continue/config.yaml`(优先)或 `config.json`(旧版) | `agentmemory connect continue` 在没有配置时创建 `config.yaml`,或更新现有 `config.json`。已有 YAML 会保留不动,以保护注释和锚点;命令打印应粘贴到 `mcpServers:` 下的配置块。Continue 的 `mcpServers` 使用数组。 |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` 写入 `context_servers` 下(这是 Zed 自己的键,不是 `mcpServers`)。远程 MCP 服务器也可以改用 `{"url": "..."}` 接入。 |
@@ -934,7 +726,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| 维护 | 手动整理的指令 | 已索引的会话历史与记忆 |
| 搜索 | 把所有内容加载进上下文 | BM25 + 向量 + 图(仅 top-K) |
-| Token 成本 | 240 条观测达 22K+ | ~1,900 tokens(减少 92%) |
| 跨代理 | 每代理一个文件 | MCP + REST(任意代理) |
| 协调 | 无 | 租约、信号、动作、例程 |
| 可观测性 | 手动读取文件 | :3113 上的实时查看器 |
@@ -947,7 +738,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1032,7 +823,7 @@ SessionStart hook fires
当向量索引已经建好时,`mem::search`(`memory_recall` 背后的实现)使用混合 BM25 + 向量排序器。没有嵌入时它使用 BM25。当图数据存在时,`smart-search` 还可以额外融合结构化的图匹配结果,即使在无密钥模式下也是如此。经验教训召回运行在一个专用的内存 BM25 索引上,而不是每次查询都扫描整个语料库。被取代的记忆版本会被排除在每条召回路径之外;版本链保留它们的历史。
-向量可以在崩溃或强制终止后存活下来。向量索引最多每隔 `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`(10 分钟)就会按桶保存一次。在此期间新增或删除的每个向量也会立刻写入状态存储中的一个小型待处理日志,下次启动时会重放这个日志而不调用嵌入提供者。每次成功保存都会清空这个日志。重放之后仍然没有向量的文档会在后台按 `AGENTMEMORY_VECTOR_BACKFILL_MAX`(500)为一批重新嵌入,直到没有遗留为止;被中断的回填会在下次启动时继续。`/agentmemory/status` 和查看器会显示待处理日志的大小和回填状态。无密钥安装不会写入任何内容。
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 开箱即用支持希腊语、西里尔语、希伯来语、阿拉伯语和带音标的拉丁文分词。对于中文 / 日语 / 韩语的记忆,安装可选的分词器(`npm install @node-rs/jieba tiny-segmenter`)来把 CJK 文本切分为词级 token;如果不安装,agentmemory 会软回退到整串分词,并在 stderr 上打印一次性提示。
@@ -1048,9 +839,9 @@ EMBEDDING_PROVIDER=local
| 提供者 | 模型 | 成本 | 备注 |
|---|---|---|---|
-| **本地(推荐选用)** | `all-MiniLM-L6-v2` | 免费 | 首次模型下载后在本地运行,召回率比仅用 BM25 高 +8pp |
+| **本地(推荐选用)** | `all-MiniLM-L6-v2` | 免费 | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | 免费层 | 100+ 种语言,768/1536/3072 维(MRL),2048-token 输入。替代 `text-embedding-004`([已弃用,2026 年 1 月 14 日下线](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | 质量最高 |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | 付费 | 针对代码优化 |
| Cohere | `embed-english-v3.0` | 免费试用 | 通用 |
| OpenRouter | 任意模型 | 视情况而定 | 多模型代理 |
diff --git a/READMEs/README.zh-TW.md b/READMEs/README.zh-TW.md
index 5a1c9029..cd1a9053 100644
--- a/READMEs/README.zh-TW.md
+++ b/READMEs/README.zh-TW.md
@@ -66,8 +66,6 @@
</p>
<p align="center">
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-recall.svg"><img src="../assets/tags/stat-recall.svg" alt="95.2% 檢索 R@5" height="38" /></picture>
- <picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tokens.svg"><img src="../assets/tags/stat-tokens.svg" alt="少 92% 的 token" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-tools.svg"><img src="../assets/tags/stat-tools.svg" alt="54 個 MCP 工具" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-hooks.svg"><img src="../assets/tags/stat-hooks.svg" alt="12 個自動 hooks" height="38" /></picture>
<picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/stat-deps.svg"><img src="../assets/tags/stat-deps.svg" alt="0 個外部資料庫" height="38" /></picture>
@@ -329,223 +327,17 @@ agentmemory --instance 1
<h2 id="benchmarks"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-benchmarks.svg"><img src="../assets/tags/section-benchmarks.svg" alt="Benchmarks" height="32" /></picture></h2>
-<table>
-<tr>
-<td width="50%">
-
-### 檢索準確率
-
-**coding-agent-life-v1**(內部語料庫,可在沙箱中重現)
-
-| 配接器 | P@5 | R@5 | Top-5 命中率 | p50 延遲 |
-|---|---|---|---|---|
-| **agentmemory hybrid** | **0.240** | **1.000** | **15 / 15** | 14 ms |
-| grep 基準線 | 0.227 | 0.967 | 15 / 15 | 0 ms |
-
-在此語料庫的 **P@5 數學上限**(0.240,見計分卡)下,達成 100% 的 Top-5 命中率。混合檢索(Hybrid)找回了每一個黃金會話;grep 在多會話時間性查詢上漏掉了 2 個黃金中的 1 個。這裡的提升在於**召回 + 時間性**,而非整體精確度。此基準測試規模小且黃金樣本稀疏;下方規模更大的 LongMemEval-S 更能看出差異。完整的依類型分解與更正說明請見:[`docs/benchmarks/2026-05-20-coding-agent-life-v1.md`](../docs/benchmarks/2026-05-20-coding-agent-life-v1.md)。
-
-**LongMemEval-S**(ICLR 2025,500 個問題)
+最新檢索結果、指標定義及重現步驟:
-| 系統 | R@5 | R@10 | MRR |
-|---|---|---|---|
-| **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
-| 僅 BM25 回退 | 86.2% | 94.6% | 71.5% |
-
-</td>
-<td width="50%">
-
-### Token 節省
-
-| 方法 | 每年 Token 數 | 每年成本 |
-|---|---|---|
-| 貼上完整上下文 | 19.5M+ | 不可行(超出上下文視窗) |
-| LLM 摘要 | ~650K | ~$500 |
-| **agentmemory** | **~170K** | **~$10** |
-| agentmemory + 本地嵌入 | ~170K | **$0** |
-
-</td>
-</tr>
-</table>
-
-> 嵌入模型:`all-MiniLM-L6-v2`(本地、免費、無需 API 金鑰)。完整報告:[`benchmark/LONGMEMEVAL.md`](../benchmark/LONGMEMEVAL.md)、[`benchmark/QUALITY.md`](../benchmark/QUALITY.md)、[`benchmark/SCALE.md`](../benchmark/SCALE.md)。競品比較:[`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md),涵蓋 agentmemory 與 mem0、Letta、Khoj、supermemory、TencentDB Agent Memory、MemPalace、Zep/Graphiti、Cognee、Hippo 的對比。
-
-**本機重現:** [`eval/README.md`](../eval/README.md),一個可插拔配接器的測試框架,涵蓋 LongMemEval `_s`(公開 500 題)+ `coding-agent-life-v1`(內部 15 會話語料庫)。Grep / 向量 / agentmemory 配接器並列評分,輸出 NDJSON,已發布的計分卡位於 [`docs/benchmarks/`](../docs/benchmarks/)。
-
-**搭配 [codegraph](https://github.com/colbymchenry/codegraph)、[Understand Anything](https://github.com/Lum1104/Understand-Anything) 和 [Graphify](https://github.com/safishamsi/graphify) 使用效果更好。** 程式碼圖索引、多代理建置流程,以及跨文件 / PDF / 圖片 / 影片的更廣泛知識圖譜。agentmemory 負責記住這些工作成果;這三個專案則點亮情境層的其餘部分。使用方式與問題路由對照表:[`docs/recipes/pairings.md`](../docs/recipes/pairings.md)。
+[0.9.31 · 2026-10-28 · strict recall-all@5](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) · [eval/README.md](../eval/README.md).
---
<h2 id="vs-competitors"><picture><source media="(prefers-color-scheme: dark)" srcset="../assets/tags/light/section-competitors.svg"><img src="../assets/tags/section-competitors.svg" alt="vs Competitors" height="32" /></picture></h2>
-<table>
-<tr>
-<th></th>
-<th>agentmemory</th>
-<th>mem0 (63K ⭐)</th>
-<th>Letta / MemGPT (24K ⭐)</th>
-<th>Khoj (36K ⭐)</th>
-<th>supermemory (29K ⭐)</th>
-<th>TencentDB Agent Memory (22K ⭐)</th>
-<th>MemPalace (54K ⭐)</th>
-<th>oracleagentmemory</th>
-<th>Hippo</th>
-<th>內建(CLAUDE.md)</th>
-</tr>
-<tr>
-<td><strong>類型</strong></td>
-<td>記憶引擎 + MCP 伺服器</td>
-<td>記憶層 API</td>
-<td>完整代理執行階段</td>
-<td>個人 AI</td>
-<td>記憶 API + 應用程式</td>
-<td>團隊記憶中樞(LLM 代理層)</td>
-<td>向量記憶(開源)</td>
-<td>記憶引擎(Oracle DB)</td>
-<td>記憶系統</td>
-<td>靜態檔案</td>
-</tr>
-<tr>
-<td><strong>檢索 R@5</strong></td>
-<td><strong>95.2%</strong></td>
-<td>68.5% (LoCoMo)</td>
-<td>83.2% (LoCoMo)</td>
-<td>N/A</td>
-<td>自行回報</td>
-<td>PersonaMem 76%(自行回報)</td>
-<td>~96.6%(自行回報)</td>
-<td>94.4%(自行回報)</td>
-<td>N/A</td>
-<td>N/A(grep)</td>
-</tr>
-<tr>
-<td><strong>自動捕捉</strong></td>
-<td>12 hooks(零人工)</td>
-<td>手動呼叫 <code>add()</code></td>
-<td>代理自行編輯</td>
-<td>手動</td>
-<td>API 端擷取</td>
-<td>代理攔截(替換 base-URL)</td>
-<td>手動</td>
-<td>API 擷取</td>
-<td>手動</td>
-<td>手動編輯</td>
-</tr>
-<tr>
-<td><strong>搜尋</strong></td>
-<td>BM25 + 向量 + 圖(RRF 融合)</td>
-<td>向量 + 圖</td>
-<td>向量(封存)</td>
-<td>語意</td>
-<td>向量 + RAG</td>
-<td>4 種資產類型(Chat / Skill / Wiki / CodeGraph)</td>
-<td>僅向量</td>
-<td>向量 + 語意</td>
-<td>衰減加權</td>
-<td>把所有內容都載入上下文</td>
-</tr>
-<tr>
-<td><strong>多代理</strong></td>
-<td>MCP + REST + 租約 + 訊號</td>
-<td>API(無協調機制)</td>
-<td>僅限於 Letta 執行階段內</td>
-<td>否</td>
-<td>否</td>
-<td>團隊角色 + 共享資產</td>
-<td>否</td>
-<td>僅限範圍內</td>
-<td>多代理共享</td>
-<td>每個代理各自的檔案</td>
-</tr>
-<tr>
-<td><strong>框架綁定</strong></td>
-<td>無(任何 MCP 客戶端)</td>
-<td>無</td>
-<td>高(必須使用 Letta)</td>
-<td>獨立</td>
-<td>無</td>
-<td>代理層攔截每一次模型呼叫</td>
-<td>無</td>
-<td>Oracle Database</td>
-<td>無</td>
-<td>每個代理各自的格式</td>
-</tr>
-<tr>
-<td><strong>外部相依元件</strong></td>
-<td>不需要獨立資料庫服務</td>
-<td>Qdrant / pgvector</td>
-<td>Postgres + 向量資料庫</td>
-<td>多種</td>
-<td>託管雲端</td>
-<td>Docker 堆疊(Core + Hub + Proxy)</td>
-<td>向量儲存</td>
-<td>Oracle AI Database</td>
-<td>無</td>
-<td>無</td>
-</tr>
-<tr>
-<td><strong>記憶生命週期</strong></td>
-<td>4 層整合 + 衰減 + 自動遺忘</td>
-<td>被動擷取</td>
-<td>由代理管理</td>
-<td>手動</td>
-<td>自動遺忘</td>
-<td>手動審查;自動路由功能開發中</td>
-<td>無</td>
-<td>未說明</td>
-<td>衰減 + 整合</td>
-<td>手動清理</td>
-</tr>
-<tr>
-<td><strong>Token 效率</strong></td>
-<td>每會話約 1,900 個 token(每年 $10)</td>
-<td>依整合方式而異</td>
-<td>核心記憶留在上下文中</td>
-<td>依情況而異</td>
-<td>雲端計價</td>
-<td>未說明</td>
-<td>無 token 預算</td>
-<td>由 LLM 驅動(依情況而異)</td>
-<td>依情況而異</td>
-<td>240 條觀測達 22K+ tokens</td>
-</tr>
-<tr>
-<td><strong>即時檢視器</strong></td>
-<td>有(連接埠 3113)</td>
-<td>雲端儀表板</td>
-<td>雲端儀表板</td>
-<td>網頁介面</td>
-<td>雲端儀表板</td>
-<td>Hub 網頁介面</td>
-<td>無</td>
-<td>無</td>
-<td>無</td>
-<td>無</td>
-</tr>
-<tr>
-<td><strong>自行架設</strong></td>
-<td>有(預設)</td>
-<td>可選</td>
-<td>可選</td>
-<td>有</td>
-<td>無(僅限雲端)</td>
-<td>有(Docker)</td>
-<td>有</td>
-<td>有(Oracle DB)</td>
-<td>有</td>
-<td>有</td>
-</tr>
-</table>
-
-<sub>基準測試說明:只有 agentmemory 的 R@5 是我們自己量測的結果(LongMemEval-S,可從 <a href="../benchmark/COMPARISON.md"><code>benchmark/COMPARISON.md</code></a> 重現)。mem0 和 Letta 的數字是它們公開發表的 LoCoMo 數據(屬於不同的資料集);MemPalace、supermemory、TencentDB(PersonaMem)和 oracleagentmemory 的數字是廠商自行回報、我們尚未獨立重現的宣稱(oracleagentmemory 的測試使用 GPT-5.5 對上 Oracle AI Database)。並列呈現僅供大致參考,並非在相同資料上的正面對比。星數為近似值,會隨時間變動。</sub>
-
-值得關注的**新進者**,詳細比較見 [`benchmark/COMPARISON.md`](../benchmark/COMPARISON.md):
-
-| 系統 | ⭐ | 切入點 |
-|--------|---|-------|
-| Zep / Graphiti | 30K | 時間性知識圖譜;已發表的時間性查詢結果最強(LongMemEval 63.8%),但圖是非同步建構,因此最新事實可能有延遲 |
-| Cognee | 30K | 文件到知識圖譜的擷取,僅支援 Python,專為結構化實體擷取而設計,而非會話捕捉 |
+產品與部署方式的最新比較及來源:
-AgentMemory 結合工作階段捕捉、可搜尋的專案歷史和本機檢視器,首次使用即可選擇免金鑰模式。
+[AgentMemory / Mem0 / Supermemory / Zep / Letta / Cognee / claude-mem](../benchmark/COMPARISON.md) · 2026-10-28.
---
@@ -778,7 +570,7 @@ Verify with `curl http://localhost:3111/agentmemory/health`. Open http://localho
啟動記憶伺服器:`npx -y @agentmemory/agentmemory@latest`
-#### 透過 `npx skills add` 使用原生 skills(50+ 代理)
+#### 透過 `npx skills add` 使用原生 skills
agentmemory 以 Claude-Code 風格的 `<dir>/SKILL.md` 格式提供 17 個 skills:9 個可呼叫的動作 skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)和 8 個代理按需載入的參考 skills(`memory-discipline`、`agentmemory-mcp-tools`、`agentmemory-rest-api`、`agentmemory-config`、`agentmemory-agents`、`agentmemory-hooks`、`agentmemory-architecture`、`write-agentmemory-skill`)。這些參考 skills 內含從原始碼產生的資料表,因此永遠不會漂移。vercel-labs 的 [`skills`](https://npmjs.com/package/skills) CLI 會把它們自動安裝到發起代理的原生 skill 目錄中,支援 50+ 個代理(Claude Code、Cursor、Cline、Continue、Droid、Warp、Codex、Antigravity、Kiro、OpenCode、Goose、Roo、Trae、Windsurf 等):
@@ -835,7 +627,7 @@ Hermes、OpenClaw、Pi 和 Oh My Pi 的連接器預設安裝原生整合和 skil
| **Antigravity IDE / 2.0** | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity --with-hooks` 會在共用的自訂目錄中安裝 MCP 和擷取 hooks。參見[Antigravity 設定與限制](../docs/plugins/antigravity.md)。 |
| **Antigravity CLI**(`agy`) | `~/.gemini/config/mcp_config.json` | `agentmemory connect antigravity-cli --with-hooks` 使用與目前 IDE 版本相同的 MCP 和 hook 設定。既有安裝應使用 `--force` 刷新;參見[升級說明](../docs/plugins/antigravity.md)。 |
| **Kiro** | `~/.kiro/settings/mcp.json` | `agentmemory connect kiro` 會寫入使用者層級的設定。工作區層級的覆寫請放在你程式碼旁的 `.kiro/settings/mcp.json`。 |
-| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` 會寫入標準的 `mcpServers` 區塊。Warp 也會自動從 `.claude/skills/` 探索 skills;一旦安裝了 Claude Code 外掛,8 個 agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`commit-context`、`commit-history`、`session-history`)就會原生出現在 Warp 的斜線指令選單中。 |
+| **Warp** | `~/.warp/.mcp.json` | `agentmemory connect warp` 會寫入標準的 `mcpServers` 區塊。Warp 也會自動從 `.claude/skills/` 探索 skills;一旦安裝了 Claude Code 外掛,9 個 agentmemory skills(`remember`、`recall`、`recap`、`handoff`、`forget`、`lesson`、`commit-context`、`commit-history`、`session-history`)就會原生出現在 Warp 的斜線指令選單中。 |
| **Cline(CLI)** | `~/.cline/mcp.json` | `agentmemory connect cline` 會寫入標準的 `mcpServers` 區塊。VS Code 擴充功能使用者:透過 Cline Settings → MCP Servers → Edit JSON 貼上同一個區塊。 |
| **Continue.dev** | `~/.continue/config.yaml`(建議)或 `config.json`(舊版) | `agentmemory connect continue` 在沒有設定時建立 `config.yaml`,或更新既有 `config.json`。既有 YAML 會保留不動,以保護註解和錨點;指令會印出應貼到 `mcpServers:` 下的設定區塊。Continue 的 `mcpServers` 使用陣列。 |
| **Zed** | `~/.config/zed/settings.json` | `agentmemory connect zed` 會寫入 `context_servers`(Zed 自己的鍵,**不是** `mcpServers`)。遠端 MCP 伺服器可改用 `{"url": "..."}` 接上。 |
@@ -926,7 +718,6 @@ Session 2: "Now add rate limiting"
|---|---|---|
| 維護 | 手動整理的指令 | 已索引的工作階段歷史與記憶 |
| 搜尋 | 把所有內容載入上下文 | BM25 + 向量 + 圖(僅 top-K) |
-| Token 成本 | 240 條觀測達 22K+ | ~1,900 個 token(少 92%) |
| 跨代理 | 每個代理各自的檔案 | MCP + REST(任何代理) |
| 協調 | 無 | 租約、訊號、動作、例程 |
| 可觀測性 | 手動讀檔 | 連接埠 3113 的即時檢視器 |
@@ -939,7 +730,7 @@ Session 2: "Now add rate limiting"
```text
PostToolUse hook fires
- -> SHA-256 dedup (5min window)
+ -> Stable event ID / durable capture inbox / offline spool
-> Privacy filter (strip secrets, API keys)
-> Store raw observation
-> Synthetic compression by default
@@ -1024,7 +815,7 @@ SessionStart hook fires
當向量索引已有資料時,`mem::search`(`memory_recall` 背後)會使用混合的 BM25 + 向量排序器。沒有嵌入時則使用 BM25。當圖資料存在時,`smart-search` 還能額外融合結構化的圖比對結果,即使在無金鑰模式下也是如此。教訓召回在專用的記憶體內 BM25 索引上執行,而非每次查詢都掃描整個語料庫。被取代的記憶版本會從每一條召回路徑中排除;版本鏈保留它們的歷史。
-向量可在崩潰或強制終止後存活。向量索引最多每 `AGENTMEMORY_INDEX_SAVE_INTERVAL_MS`(10 分鐘)以分桶方式儲存一次。期間新增或移除的每個向量,也會立即寫入狀態儲存中的一份小型待處理日誌,下次啟動時會重播它而不呼叫嵌入提供者。每次成功儲存都會清空該日誌。重播後仍沒有向量的文件,會在背景以每批 `AGENTMEMORY_VECTOR_BACKFILL_MAX`(500)筆的方式重新嵌入,直到沒有剩餘為止;被中止的回填會在下次啟動時繼續。`/agentmemory/status` 和檢視器會顯示待處理日誌大小與回填狀態。無金鑰安裝則完全不寫入任何內容。
+[0.9.31: `AGENTMEMORY_VECTOR_BACKFILL=all` / `AGENTMEMORY_VECTOR_BACKFILL_MAX`](../docs/reliability/source-and-vector-recovery.md).
BM25 開箱即可對希臘文、西里爾文、希伯來文、阿拉伯文和帶重音的拉丁文進行分詞。對於中文 / 日文 / 韓文的記憶,安裝選用的分詞器(`npm install @node-rs/jieba tiny-segmenter`),把 CJK 文字段切成詞級 token;若未安裝,agentmemory 會優雅地退回整段分詞,並在 stderr 上印出一次性提示。
@@ -1040,9 +831,9 @@ EMBEDDING_PROVIDER=local
| 提供者 | 模型 | 成本 | 備註 |
|---|---|---|---|
-| **本地(建議選用)** | `all-MiniLM-L6-v2` | 免費 | 第一次模型下載後即為裝置端運算,召回率比僅用 BM25 高 +8pp |
+| **本地(建議選用)** | `all-MiniLM-L6-v2` | 免費 | [`EMBEDDING_PROVIDER=local`](../README.md#embedding-providers) |
| Gemini | `gemini-embedding-001` | 免費額度 | 支援 100+ 種語言,768/1536/3072 維(MRL),輸入上限 2048 token。取代已棄用的 `text-embedding-004`([2026 年 1 月 14 日停用](https://ai.google.dev/gemini-api/docs/deprecations)) |
-| OpenAI | `text-embedding-3-small` | $0.02/1M | 品質最高 |
+| OpenAI | `text-embedding-3-small` | [OpenAI](https://openai.com/api/pricing/) | [text-embedding-3-small](https://platform.openai.com/docs/models/text-embedding-3-small) |
| Voyage AI | `voyage-code-3` | 付費 | 針對程式碼優化 |
| Cohere | `embed-english-v3.0` | 免費試用 | 通用型 |
| OpenRouter | 任何模型 | 依情況而異 | 多模型代理 |
diff --git a/benchmark/COMPARISON.md b/benchmark/COMPARISON.md
index 207ca0f6..23bdb749 100644
--- a/benchmark/COMPARISON.md
+++ b/benchmark/COMPARISON.md
@@ -1,181 +1,27 @@
-# AI Agent Memory: Benchmark Comparison
+# Choosing a memory system
-How agentmemory compares against other persistent memory solutions for AI coding agents.
+Checked against the linked primary sources on 2026-10-28. Products evolve; deployment, model usage and commercial plans depend on the selected configuration.
-All numbers here come from published benchmarks or public repositories. We link to primary sources wherever possible so you can reproduce.
+| Project | Main use | Deployment and integration |
+|---|---|---|
+| **[AgentMemory](../README.md)** | Shared project memory for existing coding agents | Apache-2.0 local service; automatic capture on supported integrations, MCP and REST. BM25 works without a model key; embeddings and LLM processing are optional. The CLI installs its runtime. |
+| **[Mem0](https://docs.mem0.ai/introduction)** | Memory APIs for applications and coding agents | Hosted platform or open-source library/server; SDKs, MCP and coding-agent plugins. Capture depends on the integration. [Platform and OSS differ](https://docs.mem0.ai/platform/platform-vs-oss), including graph availability. |
+| **[Supermemory](https://supermemory.ai/docs/self-hosting/overview)** | Memory and document context for applications and agents | Hosted API or local server binary, plus coding-agent plugins. The local server has embedded storage and local embeddings; its codebase is separate and non-public. Local and hosted capabilities differ. |
+| **[Zep / Graphiti](https://github.com/getzep/graphiti)** | Enterprise context and temporal knowledge graphs | Zep offers a managed or BYOC platform. Apache-2.0 Graphiti is a separate self-hosted graph framework with database and model configuration. |
+| **[Letta](https://docs.letta.com/)** | Stateful agents with managed memory | Agent harness, CLI and SDK with MemFS/shared memory; managed cloud or self-hosted runtime. It supplies the agent runtime as well as memory. |
+| **[Cognee](https://github.com/topoteretes/cognee)** | Searchable knowledge from documents, code and conversations | Open-source local library/server or Cognee Cloud, with plugins and MCP. Local extraction and embedding models support a keyless path. |
+| **[claude-mem](https://github.com/thedotmack/claude-mem)** | Capture and summarize coding sessions | Apache-2.0 project with multiple agent integrations. Model-backed summaries can use its hosted observer or a configured provider; installation and provider choices affect data flow and cost. |
----
+AgentMemory fits developers who want one local memory service shared by their existing tools, with inspectable observations, lessons, actions, source expansion and a live viewer. An application-facing API, managed enterprise platform, document graph or complete stateful-agent runtime may suit a different workload better.
-## Retrieval Accuracy (LongMemEval)
+## Compare measurements on the same terms
-[LongMemEval](https://arxiv.org/abs/2410.10813) (ICLR 2025) measures long-term memory retrieval across ~48 sessions per question on the S variant (500 questions, ~115K tokens each).
+The [current retrieval report](../docs/benchmarks/2026-10-28-retrieval-evaluation.md) measures the BM25 component and a grep baseline over identical supplied sessions. It includes strict recall-all@5, partial recall, any-hit, coverage and failures. It does not rank the products above.
-| System | Benchmark | R@5 | Notes |
-|---|---|---|---|
-| **agentmemory** (BM25 + Vector) | LongMemEval-S | **95.2%** | `all-MiniLM-L6-v2` embeddings, no API key |
-| agentmemory (BM25-only) | LongMemEval-S | 86.2% | Fallback when no embedding provider available |
-| MemPalace | LongMemEval-S | ~96.6% (self-reported) | Vendor-published number we have not independently reproduced. Vector-only with a larger embedding model and no agent-integration surface (no hooks, no MCP, no multi-agent) |
-| oracleagentmemory | LongMemEval | 94.4% (self-reported) | Vendor-published, scored with GPT-5.5 at "xhigh reasoning" and requires an Oracle AI Database. We have not reproduced it. agentmemory's 95.2% uses free local embeddings and no API key |
-| Letta / MemGPT | LoCoMo | 83.2% | Different benchmark (LoCoMo, not LongMemEval) |
-| Mem0 | LoCoMo | 68.5% | Different benchmark (LoCoMo, not LongMemEval) |
+A useful comparison must hold the dataset revision, ingestion path, retrieved unit, cutoff, model, context budget, reader and judge constant. Retrieval recall, answer accuracy, task completion and token use measure different things. Historical scores and unrelated datasets should not share a leaderboard.
-**⚠️ Apples vs oranges caveat:** only agentmemory's 95.2% is our own measured result, reproducible from the methodology below. Every other number here is the vendor's published claim, on a different benchmark or harness, that we have not independently reproduced: MemPalace and oracleagentmemory report LongMemEval (oracleagentmemory's run used GPT-5.5 at "xhigh reasoning" against an Oracle AI Database), while Letta and Mem0 publish on [LoCoMo](https://snap-stanford.github.io/LoCoMo/). Treat them as ballpark vendor claims, not a head-to-head on identical data. We'd love to run every system on the same dataset; if any maintainer wants to collaborate, open an issue.
+## Costs and data flow
-Full agentmemory methodology: [`LONGMEMEVAL.md`](LONGMEMEVAL.md)
+AgentMemory has no license fee or hosted subscription. You operate its local runtime; optional hosted models charge for their usage. Installation, updates and first-use local model downloads need network access. Memory text stays local with keyless search or local models; configured remote services receive the text they process.
----
-
-## Feature Matrix
-
-| Feature | agentmemory | mem0 | Letta/MemGPT | Khoj | supermemory | MemPalace | oracleagentmemory | Hippo |
-|---|---|---|---|---|---|---|---|---|
-| **GitHub stars** | Growing | 58K+ | 23K+ | 35K+ | 26K+ | 54K+ | PyPI (Oracle) | Trending |
-| **Type** | Memory engine + MCP server | Memory layer API | Full agent runtime | Personal AI | Memory API + app | Benchmark-focused OSS | Memory engine (Oracle DB) | Memory system |
-| **Auto-capture via hooks** | ✅ 12 lifecycle hooks | ❌ Manual `add()` | ❌ Agent self-edits | ❌ Manual | ❌ API-side extraction | ❌ Manual | ❌ API extraction | ❌ Manual |
-| **Search strategy** | BM25 + Vector + Graph | Vector + Graph | Vector (archival) | Semantic | Vector + RAG | Vector-only (large model) | Vector + semantic | Decay-weighted |
-| **Multi-agent coordination** | ✅ Leases + signals + mesh | ❌ | Runtime-internal only | ❌ | ❌ | ❌ | Scoped only (user/agent/thread) | Multi-agent shared |
-| **Framework lock-in** | None | None | High | Standalone | None (drop-in wrappers) | None | Oracle Database | None |
-| **External deps** | None | Qdrant/pgvector | Postgres + vector | Multiple | Managed cloud | Vector store | Oracle AI Database | None |
-| **Self-hostable** | ✅ default | Optional | Optional | ✅ | ❌ Cloud-only | ✅ | ✅ (needs Oracle DB) | ✅ |
-| **Knowledge graph** | ✅ Entity extraction + BFS | ✅ Mem0g variant | ❌ | Doc links | ❌ | ❌ | ❌ | ❌ |
-| **Memory decay** | ✅ Ebbinghaus + tiered | ❌ | ❌ | ❌ | ✅ Auto-forget | ❌ | ❌ | ✅ Half-lives |
-| **4-tier consolidation** | ✅ Working → episodic → semantic → procedural | ❌ | OS-inspired tiers | ❌ | ❌ | ❌ | ❌ | Episodic + semantic |
-| **Version / supersession** | ✅ Jaccard-based | Passive | ❌ | ❌ | ✅ Auto-resolve | ❌ | ❌ | ❌ |
-| **Real-time viewer** | ✅ Port 3113 | Cloud dashboard | Cloud dashboard | Web UI | Cloud dashboard | ❌ | ❌ | ❌ |
-| **Privacy filtering** | ✅ Strips secrets pre-store | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
-| **Obsidian export** | ✅ Built-in | ❌ | ❌ | Native format | ❌ | ❌ | ❌ | ❌ |
-| **Cross-agent** | ✅ MCP + REST | API calls | Within runtime | Standalone | MCP + API | Standalone | Python API | Multi-agent shared |
-| **Audit trail** | ✅ All mutations logged | ❌ | Limited | ❌ | ❌ | ❌ | ❌ | ❌ |
-| **Language SDKs** | Any (REST + MCP) | Python + TS | Python only | API | Python + TS | Python | Python only | Node |
-
----
-
-## Token Efficiency
-
-The main reason to use persistent memory at all: token cost. Here's what one year of heavy agent use looks like across approaches.
-
-| Approach | Tokens / year | Cost / year | Notes |
-|---|---|---|---|
-| Paste full history into context | 19.5M+ | Impossible | Exceeds context window after ~200 observations |
-| LLM-summarized memory (extraction-based) | ~650K | ~$500 | Lossy — summarization drops detail |
-| **agentmemory (API embeddings)** | **~170K** | **~$10** | Token-budgeted, only relevant memories injected |
-| **agentmemory (local embeddings)** | **~170K** | **$0** | `all-MiniLM-L6-v2` runs in-process |
-| supermemory | Not published | Cloud pricing | Managed API, no local token budget |
-| Mem0 | Varies by integration | Varies | Extraction-based, no token budget |
-
-**agentmemory ships with a built-in token savings calculator.** Run `npx @agentmemory/agentmemory status` after a few sessions and you'll see exactly how many tokens you've saved vs. pasting the full history.
-
----
-
-## What Each Tool Is Best At
-
-This isn't a "agentmemory wins everything" page. Different tools solve different problems.
-
-**Choose agentmemory if you want:**
-- Automatic capture with zero manual `add()` calls
-- MCP server that works across Claude Code, Cursor, Codex, Gemini CLI, etc.
-- Hybrid BM25 + vector + graph search
-- Real-time viewer to see what your agent is learning
-- Self-hostable with zero external databases
-- Privacy filtering on API keys and secrets
-- Multi-agent coordination (leases, signals, routines)
-
-**Choose Mem0 if you want:**
-- Framework-agnostic API to bolt onto an existing agent
-- Managed cloud option with a dashboard
-- Python + TypeScript SDKs for direct integration
-- Entity/relationship extraction as the primary abstraction
-
-**Choose Letta/MemGPT if you want:**
-- A full agent runtime, not just memory
-- OS-inspired memory tiers (core/archival/recall)
-- Agents that self-edit their memory via function calls
-- Long-running conversational agents (weeks/months)
-
-**Choose Khoj if you want:**
-- A personal AI second brain, not agent infrastructure
-- Document-first search over your files and the web
-- Obsidian/Notion/Emacs integrations
-- Scheduled automations and research tasks
-
-**Choose supermemory if you want:**
-- A managed memory API with server-side auto-extraction and automatic forgetting
-- Drop-in wrappers for major AI frameworks (Vercel AI, LangChain, LangGraph)
-- A hosted dashboard with no infrastructure to run yourself
-- RAG plus memory served from a single query
-
-**Choose MemPalace if you want:**
-- A simple, free, open-source vector memory store
-- To chase its self-reported retrieval benchmark (we have not reproduced it)
-- Pure retrieval over agent workflow features
-- Note: no auto-capture, no MCP, no multi-agent coordination, so you wire all integration yourself
-
-**Choose oracleagentmemory if you want:**
-- You already run on Oracle AI Database and want memory inside it
-- Enterprise Oracle stack with vector search in the same database
-- LLM-backed extraction and are fine paying for a frontier model (their benchmark used GPT-5.5)
-- Note: Python-only, Oracle Database required, no MCP, no real-time viewer
-
-**Choose Hippo if you want:**
-- Biologically-inspired memory model (decay, consolidation, sleep)
-- Multi-agent shared memory as a primary feature
-- "Forget by default, earn persistence through use" philosophy
-
-**Choose TencentDB Agent Memory if you want:**
-- Team-level shared memory: conversations, docs, and code turned into four asset types (Chat Memory, Skill, Wiki, CodeGraph) with team roles and ownership
-- Zero-integration capture via an LLM proxy (point the agent's base URL at it; no hooks or MCP required)
-- CodeGraph impact analysis (symbols, call relationships) alongside memory
-- Note: the proxy sits in front of every model call, deployment is a multi-service Docker stack (Core + Hub + Proxy), the published benchmark is PersonaMem (76%, self-reported), and automated memory routing is still in progress per their README
-
-**Choose Zep / Graphiti if you want:**
-- A temporal knowledge graph: facts carry a time dimension, so "what was true when" is a first-class query
-- The strongest published temporal-query results (LongMemEval 63.8%)
-- Note: graph construction runs in the background, so freshly ingested facts can take time to become retrievable, and per-conversation memory footprint is reported to run far above extraction-based systems
-
-**Choose Cognee if you want:**
-- Knowledge-graph construction from documents and structured data before query time
-- Entity-relationship extraction as the primary product rather than session capture
-- Note: Python-only, and built for document ingestion rather than coding-agent memory
-
----
-
-## Running Your Own Benchmarks
-
-We encourage you to measure this yourself rather than trust any README. Here's how:
-
-```bash
-# Clone the repo
-git clone https://github.com/rohitg00/agentmemory.git
-cd agentmemory && npm install
-
-# Run LongMemEval-S
-npm run bench:longmemeval
-
-# Run quality benchmark (240 observations, 20 queries)
-npm run bench:quality
-
-# Run scale benchmark
-npm run bench:scale
-
-# Run real embeddings benchmark
-npm run bench:real-embeddings
-```
-
-Results land in `benchmark/results/`. All scripts, datasets, and results are committed for reproducibility.
-
----
-
-## Corrections Welcome
-
-If you maintain one of these tools and we got a number wrong, please open an issue or PR. We'd rather have accurate numbers than convenient ones.
-
-If you want to add your tool to this comparison, open a PR with:
-1. A link to your benchmark methodology
-2. The metric and dataset you're measuring on
-3. A commit hash / version so we can reproduce
-
-**Sources:**
-- Mem0 LoCoMo benchmark: [mem0.ai blog](https://mem0.ai)
-- Letta LoCoMo benchmark: [letta.com/blog/benchmarking-ai-agent-memory](https://letta.com/blog/benchmarking-ai-agent-memory)
-- LongMemEval paper: [arxiv.org/abs/2410.10813](https://arxiv.org/abs/2410.10813)
-- LoCoMo paper: [snap-stanford.github.io/LoCoMo](https://snap-stanford.github.io/LoCoMo/)
+For commercial services, use their current plans instead of a copied monthly price: [Mem0](https://mem0.ai/pricing), [Supermemory](https://supermemory.ai/pricing), [Zep](https://www.getzep.com/pricing/), [Letta](https://docs.letta.com/), [Cognee](https://docs.cognee.ai/cognee-cloud/overview), and [claude-mem](https://github.com/thedotmack/claude-mem). Check the chosen deployment's provider, telemetry, retention and access controls before using private data.
diff --git a/benchmark/LONGMEMEVAL.md b/benchmark/LONGMEMEVAL.md
index 9b1afa41..9e0c5748 100644
--- a/benchmark/LONGMEMEVAL.md
+++ b/benchmark/LONGMEMEVAL.md
@@ -1,4 +1,6 @@
-# LongMemEval-S Benchmark Results
+# Historical LongMemEval-S Benchmark Results
+
+This report uses historical **any-hit@k** scoring and an older input pipeline. It is not a 0.9.31 strict-recall result. See the [current evaluation and receipts](../docs/benchmarks/2026-10-28-retrieval-evaluation.md); the metrics and denominators differ.
[LongMemEval](https://arxiv.org/abs/2410.10813) (ICLR 2025) is an academic benchmark for evaluating long-term memory in chat assistants. It tests 5 core abilities: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention.
diff --git a/docs/benchmarks/2026-10-28-retrieval-evaluation.md b/docs/benchmarks/2026-10-28-retrieval-evaluation.md
new file mode 100644
index 00000000..c1d4278d
--- /dev/null
+++ b/docs/benchmarks/2026-10-28-retrieval-evaluation.md
@@ -0,0 +1,57 @@
+# Retrieval evaluation, 2026-10-28
+
+AgentMemory 0.9.31 development checkout, revision `3f8784d10c2e996873d4392b85226a5d042def0e`. These runs exercise the production BM25 `SearchIndex` directly and a lexical baseline over identical supplied sessions. They do not exercise the daemon, hooks, compression, vectors, graph, persistence, or an answer-generating model.
+
+## Results
+
+| Dataset | Adapter | Strict recall-all@5 | Fractional recall@5 | Any-hit@5 | Completed / attempted |
+|---|---|---:|---:|---:|---:|
+| LongMemEval-S | BM25 component | 84.7% (398/470) | 91.7% | 96.2% (452/470) | 500/500 |
+| LongMemEval-S | grep baseline | 72.1% (339/470) | 80.8% | 88.7% (417/470) | 500/500 |
+| coding-agent-life-v1 | BM25 component | 93.3% (14/15) | 93.3% | 93.3% (14/15) | 15/15 |
+| coding-agent-life-v1 | grep baseline | 100% (15/15) | 100% | 100% (15/15) | 15/15 |
+
+All 1,030 adapter attempts completed with zero execution failures. LongMemEval has 470 answerable questions and 30 unanswerable questions. The latter are included in coverage but excluded from retrieval metrics; this run does not measure abstention quality. The small coding fixture is a development regression set, and the simpler baseline does better on it.
+
+**Strict recall-all@5** requires every gold source session among the first five distinct returned sessions. **Fractional recall@5** averages the fraction of required sources found. **Any-hit@5** needs only one. Multi-source questions can therefore have an any-hit success while still missing necessary evidence.
+
+| LongMemEval question type | Answerable | BM25 strict successes | grep strict successes |
+|---|---:|---:|---:|
+| Single-session user | 64 | 63 | 60 |
+| Single-session assistant | 56 | 55 | 50 |
+| Single-session preference | 30 | 25 | 14 |
+| Knowledge update | 72 | 70 | 66 |
+| Temporal reasoning | 127 | 99 | 85 |
+| Multi-session | 121 | 86 | 64 |
+
+Multi-session retrieval remains the weakest BM25 category: 86/121 strict successes (71.1%). A complete answer-quality evaluation still needs a fixed reader, judge, model, prompt, context budget, and matched ingestion path.
+
+## Reproduce
+
+Run from the repository root with dependencies installed:
+
+```sh
+mkdir -p benchmark/data
+curl -fL https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/98d7416c24c778c2fee6e6f3006e7a073259d48f/longmemeval_s_cleaned.json -o benchmark/data/longmemeval_s_cleaned.json
+shasum -a 256 benchmark/data/longmemeval_s_cleaned.json
+npm run eval:longmemeval -- --data benchmark/data/longmemeval_s_cleaned.json --adapters grep,bm25
+npm run eval:coding-life -- --adapters grep,bm25
+```
+
+Expected dataset SHA256: `d6f21ea9d60a0d56f34a05b609c79c88a451d2ae03597821ea3d5a9678c3a442`. The external dataset is not redistributed. The loader removes answer labels, uses opaque source IDs, retains supplied dates, and coalesces 13 identical repeated distractor IDs. Each question gets a fresh index. See the [evaluation guide](../../eval/README.md) for the input transformation and metric contract.
+
+The run used Node v25.6.1 on macOS arm64. Manifests record source, installed dependency and dataset hashes. `dirty: true` records an untracked dependency symlink; tracked source was unchanged during the run. Diagnostic query timings exclude ingestion, initialization and cleanup, have no warmup or repeated trials, and are not end-to-end latency claims.
+
+## Receipts and scope
+
+- [LongMemEval summary](receipts/2026-10-28/longmemeval/summary.json) and [manifest](receipts/2026-10-28/longmemeval/manifest.json).
+- [LongMemEval per-question scores](receipts/2026-10-28/longmemeval/scores.ndjson.gz) and [returned-source evidence](receipts/2026-10-28/longmemeval/evidence.ndjson.gz).
+- [Coding summary](receipts/2026-10-28/coding/summary.json), [manifest](receipts/2026-10-28/coding/manifest.json), [scores](receipts/2026-10-28/coding/scores.ndjson.gz), and [evidence](receipts/2026-10-28/coding/evidence.ndjson.gz).
+
+Receipts contain public fixture identifiers, content hashes and scores, without transcript text or personal memory data. Decompress the NDJSON with `gzip -dc <file>`.
+
+These results do not establish a speedup or accuracy gain over 0.9.30. The historical [hybrid LongMemEval report](../../benchmark/LONGMEMEVAL.md) used any-hit scoring and a different input pipeline; its 95.2% headline cannot be compared to strict recall-all here. The [v0.6.0 token experiment](../../benchmark/REAL-EMBEDDINGS.md) and [May coding benchmark](2026-05-20-coding-agent-life-v1.md) remain historical measurements, not 0.9.31 guarantees. No third-party memory product was run in this evaluation.
+
+## What changed in 0.9.31
+
+The [release notes](https://github.com/rohitg00/agentmemory/releases/tag/v0.9.31) document scoped saved-memory recall, incremental session summaries, paged exports, validated imports, native Pi/OMP/DeepSeek Harness integrations, Cursor workspace attribution, and compression/recovery fixes. These are implementation improvements. Lifecycle correctness and packaged startup/restart behavior are tested separately by the [evaluation suites](../../eval/README.md) and [release gate](../../scripts/release-gate/); this component run does not certify them.
diff --git a/docs/benchmarks/receipts/2026-10-28/coding/evidence.ndjson.gz b/docs/benchmarks/receipts/2026-10-28/coding/evidence.ndjson.gz
new file mode 100644
index 00000000..a5ac31f9
Binary files /dev/null and b/docs/benchmarks/receipts/2026-10-28/coding/evidence.ndjson.gz differ
diff --git a/docs/benchmarks/receipts/2026-10-28/coding/manifest.json b/docs/benchmarks/receipts/2026-10-28/coding/manifest.json
new file mode 100644
index 00000000..423dcc71
--- /dev/null
+++ b/docs/benchmarks/receipts/2026-10-28/coding/manifest.json
@@ -0,0 +1,80 @@
+{
+ "schemaVersion": 1,
+ "metricVersion": "distinct-session-retrieval-v2",
+ "inputTransform": "role-flattened; opaque source IDs; all supplied dates retained; identical repeated IDs coalesced (v1)",
+ "benchmark": "coding-agent-life-v1",
+ "scope": "supplied-session retrieval; no reader, judge, or capture-quality score",
+ "startedAt": "2026-10-28T07:28:33.253Z",
+ "source": {
+ "revision": "3f8784d10c2e996873d4392b85226a5d042def0e",
+ "dirty": true,
+ "contentSha256": "25c0fc4f610b612675600ebac66da592716081b7cf97b34c018b417f1b34520f",
+ "dependencies": {
+ "@anthropic-ai/claude-agent-sdk": "0.3.308",
+ "@anthropic-ai/sdk": "0.100.1",
+ "@clack/prompts": "1.8.1",
+ "@huggingface/transformers": "4.3.1",
+ "@iii-dev/helpers": "0.25.0",
+ "@node-rs/jieba": "2.0.4",
+ "@types/node": "26.6.5",
+ "dotenv": "18.0.8",
+ "iii-sdk": "0.25.0",
+ "json5": "2.2.3",
+ "picocolors": "1.1.1",
+ "sigstore": null,
+ "tiny-segmenter": "0.2.0",
+ "tsdown": "0.21.10",
+ "tsx": "4.23.15",
+ "typescript": "6.0.3",
+ "vitest": "4.1.11",
+ "ws": "8.22.0",
+ "yaml": "2.9.1",
+ "zod": "4.6.7"
+ },
+ "installedLockSha256": "d8a23fe5cc06b1ae740ed1eb72594e9af9f2d52231a7431fa2d509d54e9f0950",
+ "builtArtifactSha256": null,
+ "engine": null
+ },
+ "datasetFiles": [
+ {
+ "name": "sessions.json",
+ "sha256": "d32f0c3c35a35e6c5aa0ac1fd0bd92e1f413c132efe3b1262038595589df3f16"
+ },
+ {
+ "name": "queries.json",
+ "sha256": "4f36fb00ea1e0ce3ccbefea9348fc4da027201a26a1fd7b5349391bcc69a24b4"
+ }
+ ],
+ "datasetSha256": "7c37494bd6c3211e937c132b7501a594151f56337f933d8c617f79d4bf4c0701",
+ "selectedQuestionIds": [
+ "q-001",
+ "q-002",
+ "q-003",
+ "q-004",
+ "q-005",
+ "q-006",
+ "q-007",
+ "q-008",
+ "q-009",
+ "q-010",
+ "q-011",
+ "q-012",
+ "q-013",
+ "q-014",
+ "q-015"
+ ],
+ "coalescedRepeatedSources": 0,
+ "adapters": [
+ "grep",
+ "agentmemory-bm25-component"
+ ],
+ "k": 5,
+ "retrievalUnit": "top k distinct source sessions",
+ "config": {},
+ "runtime": {
+ "node": "v25.6.1",
+ "platform": "darwin",
+ "arch": "arm64"
+ },
+ "planned": 30
+}
diff --git a/docs/benchmarks/receipts/2026-10-28/coding/scores.ndjson.gz b/docs/benchmarks/receipts/2026-10-28/coding/scores.ndjson.gz
new file mode 100644
index 00000000..45a4dcc3
Binary files /dev/null and b/docs/benchmarks/receipts/2026-10-28/coding/scores.ndjson.gz differ
diff --git a/docs/benchmarks/receipts/2026-10-28/coding/summary.json b/docs/benchmarks/receipts/2026-10-28/coding/summary.json
new file mode 100644
index 00000000..33dfde9e
--- /dev/null
+++ b/docs/benchmarks/receipts/2026-10-28/coding/summary.json
@@ -0,0 +1,653 @@
+{
+ "byAdapter": {
+ "grep": {
+ "n": 15,
+ "completed": 15,
+ "failed": 0,
+ "scored": 15,
+ "answerable": {
+ "attempted": 15,
+ "completed": 15,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.24000000000000002,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 15,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.03100000000000591,
+ "latencyP95": 0.19308300000000145,
+ "ingestionP50": 0.005083000000013271
+ },
+ "agentmemory-bm25-component": {
+ "n": 15,
+ "completed": 15,
+ "failed": 0,
+ "scored": 15,
+ "answerable": {
+ "attempted": 15,
+ "completed": 15,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.22666666666666666,
+ "r": 0.9333333333333333,
+ "recallAll": 0.9333333333333333,
+ "hit": 14,
+ "strictSuccessOverAttempts": 0.9333333333333333,
+ "latencyP50": 0.08995800000002419,
+ "latencyP95": 0.4154169999999908,
+ "ingestionP50": 1.278959000000043
+ }
+ },
+ "byType": {
+ "single-session-bug": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.19308300000000145,
+ "latencyP95": 0.19308300000000145,
+ "ingestionP50": 0.07908300000002555
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.4154169999999908,
+ "latencyP95": 0.4154169999999908,
+ "ingestionP50": 3.890874999999994
+ }
+ },
+ "single-session-infra": {
+ "grep": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02533299999998917,
+ "latencyP95": 0.03741600000000744,
+ "ingestionP50": 0.0025840000000130203
+ },
+ "agentmemory-bm25-component": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1,
+ "r": 0.5,
+ "recallAll": 0.5,
+ "hit": 1,
+ "strictSuccessOverAttempts": 0.5,
+ "latencyP50": 0.08066700000000537,
+ "latencyP95": 0.14379100000002154,
+ "ingestionP50": 1.1475419999999872
+ }
+ },
+ "single-session-refactor": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.028250000000014097,
+ "latencyP95": 0.028250000000014097,
+ "ingestionP50": 0.007082999999965978
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.10045800000000327,
+ "latencyP95": 0.10045800000000327,
+ "ingestionP50": 1.338292000000024
+ }
+ },
+ "single-session-feature": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.024707999999975527,
+ "latencyP95": 0.024707999999975527,
+ "ingestionP50": 0.005375000000015007
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08070800000001555,
+ "latencyP95": 0.08070800000001555,
+ "ingestionP50": 1.6152920000000108
+ }
+ },
+ "single-session-test": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.03100000000000591,
+ "latencyP95": 0.03100000000000591,
+ "ingestionP50": 0.012625000000014097
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08995800000002419,
+ "latencyP95": 0.08995800000002419,
+ "ingestionP50": 1.11262499999998
+ }
+ },
+ "single-session-perf": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02583300000003419,
+ "latencyP95": 0.02583300000003419,
+ "ingestionP50": 0.006291999999973541
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.07287500000001046,
+ "latencyP95": 0.07287500000001046,
+ "ingestionP50": 1.278959000000043
+ }
+ },
+ "single-session-api": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.032708000000013726,
+ "latencyP95": 0.032708000000013726,
+ "ingestionP50": 0.0028750000000172804
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08158400000002075,
+ "latencyP95": 0.08158400000002075,
+ "ingestionP50": 1.1327079999999796
+ }
+ },
+ "single-session-db": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.035833000000025095,
+ "latencyP95": 0.035833000000025095,
+ "ingestionP50": 0.005083000000013271
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.13049999999998363,
+ "latencyP95": 0.13049999999998363,
+ "ingestionP50": 1.2863330000000133
+ }
+ },
+ "single-session-release": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.030915999999990618,
+ "latencyP95": 0.030915999999990618,
+ "ingestionP50": 0.002541000000007898
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.114291000000037,
+ "latencyP95": 0.114291000000037,
+ "ingestionP50": 1.3028339999999616
+ }
+ },
+ "multi-session-causal": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.04775000000000773,
+ "latencyP95": 0.04775000000000773,
+ "ingestionP50": 0.0028330000000096334
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.1052090000000021,
+ "latencyP95": 0.1052090000000021,
+ "ingestionP50": 1.1877920000000017
+ }
+ },
+ "preference": {
+ "grep": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.025458999999955267,
+ "latencyP95": 0.042124999999998636,
+ "ingestionP50": 0.0024999999999977263
+ },
+ "agentmemory-bm25-component": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.07791699999995672,
+ "latencyP95": 0.08508299999999736,
+ "ingestionP50": 1.0938330000000178
+ }
+ },
+ "multi-session-review": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.042709000000002106,
+ "latencyP95": 0.042709000000002106,
+ "ingestionP50": 0.0018750000000409273
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.12833399999999529,
+ "latencyP95": 0.12833399999999529,
+ "ingestionP50": 1.2722499999999854
+ }
+ },
+ "temporal": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02400000000000091,
+ "latencyP95": 0.02400000000000091,
+ "ingestionP50": 0.004457999999999629
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08754099999998743,
+ "latencyP95": 0.08754099999998743,
+ "ingestionP50": 1.2965829999999983
+ }
+ }
+ },
+ "planned": 30,
+ "completed": 30,
+ "failed": 0,
+ "finishedAt": "2026-10-28T07:28:33.381Z",
+ "claim": "Development evaluation only. No competitor ranking or answer-accuracy claim."
+}
diff --git a/docs/benchmarks/receipts/2026-10-28/longmemeval/evidence.ndjson.gz b/docs/benchmarks/receipts/2026-10-28/longmemeval/evidence.ndjson.gz
new file mode 100644
index 00000000..a8121930
Binary files /dev/null and b/docs/benchmarks/receipts/2026-10-28/longmemeval/evidence.ndjson.gz differ
diff --git a/docs/benchmarks/receipts/2026-10-28/longmemeval/manifest.json b/docs/benchmarks/receipts/2026-10-28/longmemeval/manifest.json
new file mode 100644
index 00000000..f9c22bc3
--- /dev/null
+++ b/docs/benchmarks/receipts/2026-10-28/longmemeval/manifest.json
@@ -0,0 +1,561 @@
+{
+ "schemaVersion": 1,
+ "metricVersion": "distinct-session-retrieval-v2",
+ "inputTransform": "role-flattened; opaque source IDs; all supplied dates retained; identical repeated IDs coalesced (v1)",
+ "benchmark": "longmemeval",
+ "scope": "supplied-session retrieval; no reader, judge, or capture-quality score",
+ "startedAt": "2026-10-28T07:27:25.386Z",
+ "source": {
+ "revision": "3f8784d10c2e996873d4392b85226a5d042def0e",
+ "dirty": true,
+ "contentSha256": "25c0fc4f610b612675600ebac66da592716081b7cf97b34c018b417f1b34520f",
+ "dependencies": {
+ "@anthropic-ai/claude-agent-sdk": "0.3.308",
+ "@anthropic-ai/sdk": "0.100.1",
+ "@clack/prompts": "1.8.1",
+ "@huggingface/transformers": "4.3.1",
+ "@iii-dev/helpers": "0.25.0",
+ "@node-rs/jieba": "2.0.4",
+ "@types/node": "26.6.5",
+ "dotenv": "18.0.8",
+ "iii-sdk": "0.25.0",
+ "json5": "2.2.3",
+ "picocolors": "1.1.1",
+ "sigstore": null,
+ "tiny-segmenter": "0.2.0",
+ "tsdown": "0.21.10",
+ "tsx": "4.23.15",
+ "typescript": "6.0.3",
+ "vitest": "4.1.11",
+ "ws": "8.22.0",
+ "yaml": "2.9.1",
+ "zod": "4.6.7"
+ },
+ "installedLockSha256": "d8a23fe5cc06b1ae740ed1eb72594e9af9f2d52231a7431fa2d509d54e9f0950",
+ "builtArtifactSha256": null,
+ "engine": null
+ },
+ "datasetFiles": [
+ {
+ "name": "longmemeval_s_cleaned.json",
+ "sha256": "d6f21ea9d60a0d56f34a05b609c79c88a451d2ae03597821ea3d5a9678c3a442"
+ }
+ ],
+ "datasetSha256": "f92fce317e7a00945830a5043fe8e877a599d0726c76e00796a1bac75902f6bd",
+ "selectedQuestionIds": [
+ "e47becba",
+ "118b2229",
+ "51a45a95",
+ "58bf7951",
+ "1e043500",
+ "c5e8278d",
+ "6ade9755",
+ "6f9b354f",
+ "58ef2f1c",
+ "f8c5f88b",
+ "5d3d2817",
+ "7527f7e2",
+ "c960da58",
+ "3b6f954b",
+ "726462e0",
+ "94f70d80",
+ "66f24dbb",
+ "ad7109d1",
+ "af8d2e46",
+ "dccbc061",
+ "c8c3f81d",
+ "8ebdbe50",
+ "6b168ec8",
+ "75499fd8",
+ "21436231",
+ "95bcc1c8",
+ "0862e8bf",
+ "853b0a1d",
+ "a06e4cfe",
+ "37d43f65",
+ "b86304ba",
+ "d52b4f67",
+ "25e5aa4f",
+ "caf9ead2",
+ "8550ddae",
+ "60d45044",
+ "3f1e9474",
+ "86b68151",
+ "577d4d32",
+ "ec81a493",
+ "15745da0",
+ "e01b8e2f",
+ "bc8a6e93",
+ "ccb36322",
+ "001be529",
+ "b320f3f8",
+ "19b5f2b3",
+ "4fd1909e",
+ "545bd2b5",
+ "8a137a7f",
+ "76d63226",
+ "86f00804",
+ "8e9d538c",
+ "311778f1",
+ "c19f7a0b",
+ "4100d0a0",
+ "29f2956b",
+ "1faac195",
+ "faba32e5",
+ "f4f1d8a4",
+ "c14c00dd",
+ "36580ce8",
+ "3d86fd0a",
+ "a82c026e",
+ "0862e8bf_abs",
+ "15745da0_abs",
+ "bc8a6e93_abs",
+ "19b5f2b3_abs",
+ "29f2956b_abs",
+ "f4f1d8a4_abs",
+ "0a995998",
+ "6d550036",
+ "gpt4_59c863d7",
+ "b5ef892d",
+ "e831120c",
+ "3a704032",
+ "gpt4_d84a3211",
+ "aae3761f",
+ "gpt4_f2262a51",
+ "dd2973ad",
+ "c4a1ceb8",
+ "gpt4_a56e767c",
+ "6cb6f249",
+ "46a3abf7",
+ "36b9f61e",
+ "28dc39ac",
+ "gpt4_2f8be40d",
+ "2e6d26dc",
+ "gpt4_15e38248",
+ "88432d0a",
+ "80ec1f4f",
+ "d23cf73b",
+ "gpt4_7fce9456",
+ "d682f1a2",
+ "7024f17c",
+ "gpt4_5501fe77",
+ "gpt4_2ba83207",
+ "2318644b",
+ "2ce6a0f2",
+ "gpt4_d12ceb0e",
+ "00ca467f",
+ "b3c15d39",
+ "gpt4_31ff4165",
+ "eeda8a6d",
+ "2788b940",
+ "60bf93ed",
+ "9d25d4e0",
+ "129d1232",
+ "60472f9c",
+ "gpt4_194be4b3",
+ "a9f6b44c",
+ "d851d5ba",
+ "5a7937c8",
+ "gpt4_ab202e7f",
+ "gpt4_e05b82a6",
+ "gpt4_731e37d7",
+ "edced276",
+ "10d9b85a",
+ "e3038f8c",
+ "2b8f3739",
+ "1a8a66a6",
+ "c2ac3c61",
+ "bf659f65",
+ "gpt4_372c3eed",
+ "gpt4_2f91af09",
+ "81507db6",
+ "88432d0a_abs",
+ "80ec1f4f_abs",
+ "eeda8a6d_abs",
+ "60bf93ed_abs",
+ "edced276_abs",
+ "gpt4_372c3eed_abs",
+ "8a2466db",
+ "06878be2",
+ "75832dbd",
+ "0edc2aef",
+ "35a27287",
+ "32260d93",
+ "195a1a1b",
+ "afdc33df",
+ "caf03d32",
+ "54026fce",
+ "06f04340",
+ "6b7dfb22",
+ "1a1907b4",
+ "09d032c9",
+ "38146c39",
+ "d24813b1",
+ "57f827a0",
+ "95228167",
+ "505af2f5",
+ "75f70248",
+ "d6233ab6",
+ "1da05512",
+ "fca70973",
+ "b6025781",
+ "a89d7624",
+ "b0479f84",
+ "1d4e3b97",
+ "07b6f563",
+ "1c0ddc50",
+ "0a34ad58",
+ "d3ab962e",
+ "2311e44b",
+ "cc06de0d",
+ "a11281a2",
+ "4f54b7c9",
+ "85fa3a3f",
+ "9aaed6a3",
+ "1f2b8d4f",
+ "e6041065",
+ "51c32626",
+ "d905b33f",
+ "7405e8b1",
+ "f35224e0",
+ "6456829e",
+ "a4996e51",
+ "3c1045c8",
+ "60036106",
+ "681a1674",
+ "e25c3b8d",
+ "4adc0475",
+ "4bc144e2",
+ "ef66a6e5",
+ "5025383b",
+ "a1cc6108",
+ "9ee3ecd6",
+ "3fdac837",
+ "91b15a6e",
+ "27016adc",
+ "720133ac",
+ "77eafa52",
+ "8979f9ec",
+ "0100672e",
+ "a96c20ee",
+ "92a0aa75",
+ "3fe836c9",
+ "1c549ce4",
+ "6c49646a",
+ "1192316e",
+ "0ea62687",
+ "67e0d0f2",
+ "bb7c3b45",
+ "ba358f49",
+ "61f8c8f8",
+ "60159905",
+ "ef9cf60a",
+ "73d42213",
+ "bc149d6b",
+ "099778bb",
+ "09ba9854",
+ "d6062bb9",
+ "157a136e",
+ "c18a7dc8",
+ "a3332713",
+ "55241a1f",
+ "a08a253f",
+ "f0e564bc",
+ "078150f1",
+ "8cf4d046",
+ "a346bb18",
+ "37f165cf",
+ "8e91e7d9",
+ "87f22b4a",
+ "e56a43b9",
+ "efc3f7c2",
+ "21d02d0d",
+ "2311e44b_abs",
+ "6456829e_abs",
+ "e5ba910e_abs",
+ "a96c20ee_abs",
+ "ba358f49_abs",
+ "09ba9854_abs",
+ "gpt4_59149c77",
+ "gpt4_f49edff3",
+ "71017276",
+ "b46e15ed",
+ "gpt4_fa19884c",
+ "0bc8ad92",
+ "af082822",
+ "gpt4_4929293a",
+ "gpt4_b5700ca9",
+ "9a707b81",
+ "gpt4_1d4ab0c9",
+ "gpt4_e072b769",
+ "0db4c65d",
+ "gpt4_1d80365e",
+ "gpt4_7f6b06db",
+ "gpt4_6dc9b45b",
+ "gpt4_8279ba02",
+ "gpt4_18c2b244",
+ "gpt4_a1b77f9c",
+ "gpt4_1916e0ea",
+ "gpt4_7a0daae1",
+ "gpt4_468eb063",
+ "gpt4_7abb270c",
+ "gpt4_1e4a8aeb",
+ "gpt4_4fc4f797",
+ "4dfccbf7",
+ "gpt4_61e13b3c",
+ "gpt4_45189cb4",
+ "2ebe6c90",
+ "gpt4_e061b84f",
+ "370a8ff4",
+ "gpt4_d6585ce8",
+ "gpt4_4ef30696",
+ "gpt4_ec93e27f",
+ "6e984301",
+ "8077ef71",
+ "gpt4_f420262c",
+ "gpt4_8e165409",
+ "gpt4_74aed68e",
+ "bcbe585f",
+ "gpt4_21adecb5",
+ "5e1b23de",
+ "gpt4_98f46fc6",
+ "gpt4_af6db32f",
+ "eac54adc",
+ "gpt4_7ddcf75f",
+ "gpt4_a2d1d1f6",
+ "gpt4_85da3956",
+ "gpt4_b0863698",
+ "gpt4_68e94287",
+ "gpt4_e414231e",
+ "gpt4_7ca326fa",
+ "gpt4_7bc6cf22",
+ "2ebe6c92",
+ "gpt4_e061b84g",
+ "71017277",
+ "b46e15ee",
+ "gpt4_d6585ce9",
+ "gpt4_1e4a8aec",
+ "gpt4_f420262d",
+ "gpt4_59149c78",
+ "gpt4_e414231f",
+ "gpt4_4929293b",
+ "gpt4_468eb064",
+ "gpt4_fa19884d",
+ "9a707b82",
+ "eac54add",
+ "4dfccbf8",
+ "0bc8ad93",
+ "6e984302",
+ "gpt4_8279ba03",
+ "gpt4_b5700ca0",
+ "gpt4_68e94288",
+ "gpt4_2655b836",
+ "gpt4_2487a7cb",
+ "gpt4_76048e76",
+ "gpt4_2312f94c",
+ "0bb5a684",
+ "08f4fc43",
+ "2c63a862",
+ "gpt4_385a5000",
+ "2a1811e2",
+ "bbf86515",
+ "gpt4_5dcc0aab",
+ "gpt4_0b2f1d21",
+ "f0853d11",
+ "gpt4_6ed717ea",
+ "gpt4_70e84552",
+ "a3838d2b",
+ "gpt4_93159ced",
+ "gpt4_2d58bcd6",
+ "gpt4_65aabe59",
+ "982b5123",
+ "b9cfe692",
+ "gpt4_4edbafa2",
+ "c8090214",
+ "gpt4_483dd43c",
+ "e4e14d04",
+ "c9f37c46",
+ "gpt4_2c50253f",
+ "dcfa8644",
+ "gpt4_b4a80587",
+ "gpt4_9a159967",
+ "cc6d1ec1",
+ "gpt4_8c8961ae",
+ "gpt4_d9af6064",
+ "gpt4_7de946e7",
+ "d01c6aa8",
+ "993da5e2",
+ "a3045048",
+ "gpt4_d31cdae3",
+ "gpt4_cd90e484",
+ "gpt4_88806d6e",
+ "gpt4_4cd9eba1",
+ "gpt4_93f6379c",
+ "b29f3365",
+ "gpt4_2f56ae70",
+ "6613b389",
+ "gpt4_78cf46a3",
+ "gpt4_0a05b494",
+ "gpt4_1a1dc16d",
+ "gpt4_2f584639",
+ "gpt4_213fd887",
+ "gpt4_5438fa52",
+ "gpt4_c27434e8",
+ "gpt4_fe651585",
+ "8c18457d",
+ "gpt4_70e84552_abs",
+ "gpt4_93159ced_abs",
+ "982b5123_abs",
+ "c8090214_abs",
+ "gpt4_c27434e8_abs",
+ "gpt4_fe651585_abs",
+ "6a1eabeb",
+ "6aeb4375",
+ "830ce83f",
+ "852ce960",
+ "945e3d21",
+ "d7c942c3",
+ "71315a70",
+ "89941a93",
+ "ce6d2d27",
+ "9ea5eabc",
+ "07741c44",
+ "a1eacc2a",
+ "184da446",
+ "031748ae",
+ "4d6b87c8",
+ "0f05491a",
+ "08e075c7",
+ "f9e8c073",
+ "41698283",
+ "2698e78f",
+ "b6019101",
+ "45dc21b6",
+ "5a4f22c0",
+ "6071bd76",
+ "e493bb7c",
+ "618f13b2",
+ "72e3ee87",
+ "c4ea545c",
+ "01493427",
+ "6a27ffc2",
+ "2133c1b5",
+ "18bc8abd",
+ "db467c8c",
+ "7a87bd0c",
+ "e61a7584",
+ "1cea1afa",
+ "ed4ddc30",
+ "8fb83627",
+ "b01defab",
+ "22d2cb42",
+ "0e4e4c46",
+ "4b24c848",
+ "7e974930",
+ "603deb26",
+ "59524333",
+ "5831f84d",
+ "eace081b",
+ "affe2881",
+ "50635ada",
+ "e66b632c",
+ "0ddfec37",
+ "f685340e",
+ "cc5ded98",
+ "dfde3500",
+ "69fee5aa",
+ "7401057b",
+ "cf22b7bf",
+ "a2f3aa27",
+ "c7dc5443",
+ "06db6396",
+ "3ba21379",
+ "9bbe84a2",
+ "10e09553",
+ "dad224aa",
+ "ba61f0b9",
+ "42ec0761",
+ "5c40ec5b",
+ "c6853660",
+ "26bdc477",
+ "0977f2af",
+ "6aeb4375_abs",
+ "031748ae_abs",
+ "2698e78f_abs",
+ "2133c1b5_abs",
+ "0ddfec37_abs",
+ "f685340e_abs",
+ "89941a94",
+ "07741c45",
+ "7161e7e2",
+ "c4f10528",
+ "89527b6b",
+ "e9327a54",
+ "4c36ccef",
+ "6ae235be",
+ "7e00a6cb",
+ "1903aded",
+ "ceb54acb",
+ "f523d9fe",
+ "0e5e2d1a",
+ "fea54f57",
+ "cc539528",
+ "dc439ea3",
+ "18dcd5a5",
+ "488d3006",
+ "58470ed2",
+ "8cf51dda",
+ "1d4da289",
+ "8464fc84",
+ "8aef76bc",
+ "71a3fd6b",
+ "2bf43736",
+ "70b3e69b",
+ "8752c811",
+ "3249768e",
+ "1b9b7252",
+ "1568498a",
+ "6222b6eb",
+ "e8a79c70",
+ "d596882b",
+ "e3fc4d6e",
+ "51b23612",
+ "3e321797",
+ "e982271f",
+ "352ab8bd",
+ "fca762bc",
+ "7a8d0b71",
+ "a40e080f",
+ "8b9d4367",
+ "5809eb10",
+ "41275add",
+ "4388e9dd",
+ "4baee567",
+ "561fabcd",
+ "b759caee",
+ "ac031881",
+ "28bcfaac",
+ "16c90bf4",
+ "c8f1aeed",
+ "eaca4986",
+ "c7cf7dfd",
+ "e48988bc",
+ "1de5cff2",
+ "65240037",
+ "778164c6"
+ ],
+ "coalescedRepeatedSources": 13,
+ "adapters": [
+ "grep",
+ "agentmemory-bm25-component"
+ ],
+ "k": 5,
+ "retrievalUnit": "top k distinct source sessions",
+ "config": {},
+ "runtime": {
+ "node": "v25.6.1",
+ "platform": "darwin",
+ "arch": "arm64"
+ },
+ "planned": 1000
+}
diff --git a/docs/benchmarks/receipts/2026-10-28/longmemeval/scores.ndjson.gz b/docs/benchmarks/receipts/2026-10-28/longmemeval/scores.ndjson.gz
new file mode 100644
index 00000000..69bc15c2
Binary files /dev/null and b/docs/benchmarks/receipts/2026-10-28/longmemeval/scores.ndjson.gz differ
diff --git a/docs/benchmarks/receipts/2026-10-28/longmemeval/summary.json b/docs/benchmarks/receipts/2026-10-28/longmemeval/summary.json
new file mode 100644
index 00000000..ce8de913
--- /dev/null
+++ b/docs/benchmarks/receipts/2026-10-28/longmemeval/summary.json
@@ -0,0 +1,331 @@
+{
+ "byAdapter": {
+ "grep": {
+ "n": 500,
+ "completed": 500,
+ "failed": 0,
+ "scored": 470,
+ "answerable": {
+ "attempted": 470,
+ "completed": 470,
+ "failed": 0
+ },
+ "unanswerable": 30,
+ "unanswerableCompleted": 30,
+ "unanswerableFailed": 0,
+ "p": 0.2953191489361707,
+ "r": 0.8081914893617022,
+ "recallAll": 0.7212765957446808,
+ "hit": 417,
+ "strictSuccessOverAttempts": 0.7212765957446808,
+ "latencyP50": 2.352957999999944,
+ "latencyP95": 5.313833000000159,
+ "ingestionP50": 0.002499999999827196
+ },
+ "agentmemory-bm25-component": {
+ "n": 500,
+ "completed": 500,
+ "failed": 0,
+ "scored": 470,
+ "answerable": {
+ "attempted": 470,
+ "completed": 470,
+ "failed": 0
+ },
+ "unanswerable": 30,
+ "unanswerableCompleted": 30,
+ "unanswerableFailed": 0,
+ "p": 0.3370212765957453,
+ "r": 0.9171985815602839,
+ "recallAll": 0.8468085106382979,
+ "hit": 452,
+ "strictSuccessOverAttempts": 0.8468085106382979,
+ "latencyP50": 1.752375000000029,
+ "latencyP95": 2.0167920000021695,
+ "ingestionP50": 128.29283299999952
+ }
+ },
+ "byType": {
+ "single-session-user": {
+ "grep": {
+ "n": 70,
+ "completed": 70,
+ "failed": 0,
+ "scored": 64,
+ "answerable": {
+ "attempted": 64,
+ "completed": 64,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.18749999999999983,
+ "r": 0.9375,
+ "recallAll": 0.9375,
+ "hit": 60,
+ "strictSuccessOverAttempts": 0.9375,
+ "latencyP50": 1.6280829999998332,
+ "latencyP95": 2.776290999999901,
+ "ingestionP50": 0.0018749999999272404
+ },
+ "agentmemory-bm25-component": {
+ "n": 70,
+ "completed": 70,
+ "failed": 0,
+ "scored": 64,
+ "answerable": {
+ "attempted": 64,
+ "completed": 64,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.1968749999999998,
+ "r": 0.984375,
+ "recallAll": 0.984375,
+ "hit": 63,
+ "strictSuccessOverAttempts": 0.984375,
+ "latencyP50": 1.7266249999993306,
+ "latencyP95": 1.9502500000007785,
+ "ingestionP50": 127.6830000000009
+ }
+ },
+ "multi-session": {
+ "grep": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 121,
+ "answerable": {
+ "attempted": 121,
+ "completed": 121,
+ "failed": 0
+ },
+ "unanswerable": 12,
+ "unanswerableCompleted": 12,
+ "unanswerableFailed": 0,
+ "p": 0.37355371900826406,
+ "r": 0.7450413223140495,
+ "recallAll": 0.5289256198347108,
+ "hit": 114,
+ "strictSuccessOverAttempts": 0.5289256198347108,
+ "latencyP50": 2.219832999999653,
+ "latencyP95": 3.6175829999997404,
+ "ingestionP50": 0.0025840000002972374
+ },
+ "agentmemory-bm25-component": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 121,
+ "answerable": {
+ "attempted": 121,
+ "completed": 121,
+ "failed": 0
+ },
+ "unanswerable": 12,
+ "unanswerableCompleted": 12,
+ "unanswerableFailed": 0,
+ "p": 0.44132231404958633,
+ "r": 0.8701101928374656,
+ "recallAll": 0.7107438016528925,
+ "hit": 117,
+ "strictSuccessOverAttempts": 0.7107438016528925,
+ "latencyP50": 1.7304590000021562,
+ "latencyP95": 1.9607500000001892,
+ "ingestionP50": 128.9686669999992
+ }
+ },
+ "single-session-preference": {
+ "grep": {
+ "n": 30,
+ "completed": 30,
+ "failed": 0,
+ "scored": 30,
+ "answerable": {
+ "attempted": 30,
+ "completed": 30,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.09333333333333334,
+ "r": 0.4666666666666667,
+ "recallAll": 0.4666666666666667,
+ "hit": 14,
+ "strictSuccessOverAttempts": 0.4666666666666667,
+ "latencyP50": 2.378375000000233,
+ "latencyP95": 3.53387500000008,
+ "ingestionP50": 0.002540999999837368
+ },
+ "agentmemory-bm25-component": {
+ "n": 30,
+ "completed": 30,
+ "failed": 0,
+ "scored": 30,
+ "answerable": {
+ "attempted": 30,
+ "completed": 30,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1666666666666667,
+ "r": 0.8333333333333334,
+ "recallAll": 0.8333333333333334,
+ "hit": 25,
+ "strictSuccessOverAttempts": 0.8333333333333334,
+ "latencyP50": 1.8002090000009048,
+ "latencyP95": 2.076791999999841,
+ "ingestionP50": 130.78970800000025
+ }
+ },
+ "temporal-reasoning": {
+ "grep": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 127,
+ "answerable": {
+ "attempted": 127,
+ "completed": 127,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.32755905511810973,
+ "r": 0.7692913385826772,
+ "recallAll": 0.6692913385826772,
+ "hit": 109,
+ "strictSuccessOverAttempts": 0.6692913385826772,
+ "latencyP50": 2.7004580000002534,
+ "latencyP95": 4.363542000000052,
+ "ingestionP50": 0.003750000000309228
+ },
+ "agentmemory-bm25-component": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 127,
+ "answerable": {
+ "attempted": 127,
+ "completed": 127,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.37795275590551125,
+ "r": 0.8803149606299214,
+ "recallAll": 0.7795275590551181,
+ "hit": 120,
+ "strictSuccessOverAttempts": 0.7795275590551181,
+ "latencyP50": 1.7312919999967562,
+ "latencyP95": 1.9771249999976135,
+ "ingestionP50": 129.1073749999996
+ }
+ },
+ "knowledge-update": {
+ "grep": {
+ "n": 78,
+ "completed": 78,
+ "failed": 0,
+ "scored": 72,
+ "answerable": {
+ "attempted": 72,
+ "completed": 72,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.3777777777777773,
+ "r": 0.9444444444444444,
+ "recallAll": 0.9166666666666666,
+ "hit": 70,
+ "strictSuccessOverAttempts": 0.9166666666666666,
+ "latencyP50": 2.0640830000002097,
+ "latencyP95": 3.4398329999999078,
+ "ingestionP50": 0.0018749999999272404
+ },
+ "agentmemory-bm25-component": {
+ "n": 78,
+ "completed": 78,
+ "failed": 0,
+ "scored": 72,
+ "answerable": {
+ "attempted": 72,
+ "completed": 72,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.394444444444444,
+ "r": 0.9861111111111112,
+ "recallAll": 0.9722222222222222,
+ "hit": 72,
+ "strictSuccessOverAttempts": 0.9722222222222222,
+ "latencyP50": 1.7199999999938882,
+ "latencyP95": 1.9851669999989099,
+ "ingestionP50": 127.13654200000019
+ }
+ },
+ "single-session-assistant": {
+ "grep": {
+ "n": 56,
+ "completed": 56,
+ "failed": 0,
+ "scored": 56,
+ "answerable": {
+ "attempted": 56,
+ "completed": 56,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.17857142857142852,
+ "r": 0.8928571428571429,
+ "recallAll": 0.8928571428571429,
+ "hit": 50,
+ "strictSuccessOverAttempts": 0.8928571428571429,
+ "latencyP50": 5.13412500000004,
+ "latencyP95": 7.704250000000229,
+ "ingestionP50": 0.0025000000000545697
+ },
+ "agentmemory-bm25-component": {
+ "n": 56,
+ "completed": 56,
+ "failed": 0,
+ "scored": 56,
+ "answerable": {
+ "attempted": 56,
+ "completed": 56,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1964285714285713,
+ "r": 0.9821428571428571,
+ "recallAll": 0.9821428571428571,
+ "hit": 55,
+ "strictSuccessOverAttempts": 0.9821428571428571,
+ "latencyP50": 1.8950000000040745,
+ "latencyP95": 2.241667000002053,
+ "ingestionP50": 127.22987499999726
+ }
+ }
+ },
+ "planned": 1000,
+ "completed": 1000,
+ "failed": 0,
+ "finishedAt": "2026-10-28T07:28:32.771Z",
+ "claim": "Development evaluation only. No competitor ranking or answer-accuracy claim."
+}
diff --git a/eval/README.md b/eval/README.md
index e09121e9..621fddc8 100644
--- a/eval/README.md
+++ b/eval/README.md
@@ -1,6 +1,6 @@
# AgentMemory evaluations
-Development evaluations for retrieval and persistent-memory behavior. These are separate from end-to-end answer accuracy, agent task success, and competitor rankings. Current measured results and remaining failures are in [the baseline report](../docs/benchmarks/2026-10-09-evaluation-baseline.md).
+Development evaluations for retrieval and persistent-memory behavior. These are separate from end-to-end answer accuracy, agent task success, and competitor rankings. Current component results are in the [2026-10-28 retrieval report](../docs/benchmarks/2026-10-28-retrieval-evaluation.md). The [2026-10-09 baseline](../docs/benchmarks/2026-10-09-evaluation-baseline.md) preserves the earlier runtime results and failures.
## Run locally
diff --git a/src/cli.ts b/src/cli.ts
index bd32b08e..ee794db4 100644
--- a/src/cli.ts
+++ b/src/cli.ts
@@ -40,6 +40,7 @@ const c = {
import { generateId } from "./state/schema.js";
import {
buildDiagnostics,
+ canAutoFix,
dryRunPlan,
parseEnvFile,
type Diagnostic,
@@ -2041,6 +2042,12 @@ function buildDoctorEffects(): DoctorEffects {
return state?.kind === "compose" ? state.binPath : privateIiiPath();
},
iiiBinaryVersion: (binPath: string) => iiiBinVersion(binPath),
+ externalEngineRunning: async () => {
+ const state = readEngineState();
+ const remoteTarget = shouldUseAgentmemoryUrl() && process.env["AGENTMEMORY_URL"];
+ const managed = state?.kind === "compose" && state.restPort === getRestPort() && !remoteTarget;
+ return !managed && await isEngineRunning();
+ },
viewerReachable: async (timeoutMs = 2000) => {
try {
await discoverViewerPort();
@@ -2340,12 +2347,15 @@ async function runDoctor() {
}
failed++;
p.log.warn(`${d.id} ✗ ${status.detail ?? ""}`.trim());
- p.log.info(`why: ${d.fixPreview}`);
- if (d.manualOnly) {
- p.log.info(`(manual fix only — see "${d.id}" docs)`);
+ if (!canAutoFix(d, status)) {
+ p.log.info(`manual fix: ${status.manualFix ?? d.fixPreview}`);
+ skipped++;
+ continue;
}
+ p.log.info(`why: ${d.fixPreview}`);
+
if (applyAll) {
const r = await applyFixWithReport(d, ctx, false);
if (r.ok) fixed++;
diff --git a/src/cli/doctor-diagnostics.ts b/src/cli/doctor-diagnostics.ts
index 2af80414..d6b7e772 100644
--- a/src/cli/doctor-diagnostics.ts
+++ b/src/cli/doctor-diagnostics.ts
@@ -22,6 +22,7 @@ export type DiagnosticStatus = {
ok: boolean;
/** Short status detail (one line). Shown alongside the check name. */
detail?: string;
+ manualFix?: string;
};
export type DiagnosticFixResult = {
@@ -61,6 +62,10 @@ export type Diagnostic = {
manualOnly?: boolean;
};
+export function canAutoFix(d: Diagnostic, status?: DiagnosticStatus): boolean {
+ return !d.manualOnly && !status?.manualFix;
+}
+
// Diagnostic ids are stable for testing and machine-readable doctor output.
export const DIAGNOSTIC_IDS = [
"env-missing",
@@ -145,6 +150,7 @@ export type DoctorEffects = {
localBinIiiPath: () => string;
/** Run `iii --version`; null if it fails. */
iiiBinaryVersion: (binPath: string) => string | null;
+ externalEngineRunning: () => Promise<boolean>;
/** Probe the viewer URL; true if it returns OK within timeoutMs. */
viewerReachable: (timeoutMs?: number) => Promise<boolean>;
/** Run init logic (copies .env.example). */
@@ -221,25 +227,51 @@ export function buildDiagnostics(effects: DoctorEffects): Diagnostic[] {
},
{
id: "engine-version-mismatch",
- message: "iii binary on PATH doesn't match the version agentmemory pins to.",
+ message: "No iii binary matches the version agentmemory pins to.",
fixPreview:
"Reinstall the pinned iii-engine and restart the managed Worker Compose project.",
moreInfo:
"agentmemory manages its pinned iii engine and workers through Worker Compose; " +
"a foreign process on PATH is not the instance binary. " +
"Running a mismatched binary surfaces as EPIPE " +
- "reconnect loops and empty search results.",
+ "reconnect loops and empty search results. Before an instance is configured, " +
+ "a matching private install at ~/.agentmemory/bin/iii can satisfy the pin.",
check: async (ctx) => {
+ if (await effects.externalEngineRunning()) {
+ return {
+ ok: false,
+ detail: "external engine version is unknown",
+ manualFix: "Verify the external engine version, or stop it before starting a managed instance.",
+ };
+ }
+ const privateVersion = effects.iiiBinaryVersion(effects.localBinIiiPath());
const bin = effects.findIiiBinary();
- if (!bin) return { ok: false, detail: "iii not on PATH" };
- const v = effects.iiiBinaryVersion(bin);
- if (!v) return { ok: false, detail: "iii on PATH but --version failed" };
+ const pathVersion = bin ? effects.iiiBinaryVersion(bin) : null;
+ if (privateVersion === ctx.pinnedVersion) {
+ const note =
+ pathVersion && pathVersion !== ctx.pinnedVersion
+ ? `; PATH iii is ${pathVersion}; the pinned private install is available for startup`
+ : "";
+ return {
+ ok: true,
+ detail: `private install ${privateVersion} (pinned ${ctx.pinnedVersion})${note}`,
+ };
+ }
+ if (!bin) {
+ return { ok: false, detail: "iii not on PATH and no private install" };
+ }
+ if (!pathVersion) {
+ return { ok: false, detail: "iii on PATH but --version failed" };
+ }
return {
- ok: v === ctx.pinnedVersion,
- detail: `${v} (pinned ${ctx.pinnedVersion})`,
+ ok: pathVersion === ctx.pinnedVersion,
+ detail: `${pathVersion} (pinned ${ctx.pinnedVersion})`,
};
},
fix: async () => {
+ if (await effects.externalEngineRunning()) {
+ return { ok: false, message: "Verify or stop the external engine manually." };
+ }
const r = await effects.runIiiInstaller();
if (!r.ok) return r;
// Best-effort restart: stop then start.
@@ -315,7 +347,7 @@ export function buildDiagnostics(effects: DoctorEffects): Diagnostic[] {
{
id: "iii-on-path-not-local-bin",
message:
- "iii is on PATH but not at agentmemory's private install path.",
+ "iii is on PATH but agentmemory's private install is absent or mismatched.",
fixPreview:
"Install the pinned version to ~/.agentmemory/bin — won't touch your PATH.",
moreInfo:
@@ -324,14 +356,19 @@ export function buildDiagnostics(effects: DoctorEffects): Diagnostic[] {
"When agentmemory needs the pin and PATH doesn't have it, it falls back to the " +
"private install. If neither exists, run the installer.",
manualOnly: true,
- check: async () => {
+ check: async (ctx) => {
const bin = effects.findIiiBinary();
if (!bin) return { ok: true, detail: "iii not on PATH (handled elsewhere)" };
const localBin = effects.localBinIiiPath();
- return {
- ok: bin === localBin,
- detail: bin === localBin ? undefined : `iii at: ${bin}`,
- };
+ if (bin === localBin) return { ok: true };
+ const privateVersion = effects.iiiBinaryVersion(localBin);
+ if (privateVersion === ctx.pinnedVersion) {
+ return {
+ ok: true,
+ detail: `private install pinned; PATH iii at ${bin} stays untouched`,
+ };
+ }
+ return { ok: false, detail: `iii at: ${bin}` };
},
fix: async () =>
effects.runIiiInstaller().then((r) => ({
@@ -376,7 +413,7 @@ export function dryRunPlan(
if (status.ok) continue;
n++;
lines.push(`${n}. [${diagnostic.id}] ${diagnostic.message}`);
- lines.push(` would fix: ${diagnostic.fixPreview}`);
+ lines.push(` ${canAutoFix(diagnostic, status) ? "would fix" : "manual fix"}: ${status.manualFix ?? diagnostic.fixPreview}`);
if (status.detail) lines.push(` detail: ${status.detail}`);
}
if (lines.length === 0) {
diff --git a/test/cli-doctor-fixes.test.ts b/test/cli-doctor-fixes.test.ts
index 41483517..333d1985 100644
--- a/test/cli-doctor-fixes.test.ts
+++ b/test/cli-doctor-fixes.test.ts
@@ -5,10 +5,11 @@
// formatting. The full interactive prompt loop lives in src/cli.ts and is
// driven by clack — exercising it would require a TTY and is out of scope.
-import { describe, it, expect } from "vitest";
+import { describe, it, expect, vi } from "vitest";
import { III_PINNED_VERSION } from "../src/version.js";
import {
buildDiagnostics,
+ canAutoFix,
DIAGNOSTIC_IDS,
dryRunPlan,
parseEnvFile,
@@ -40,6 +41,7 @@ function stubEffects(overrides: Partial<DoctorEffects> = {}): DoctorEffects {
findIiiBinary: () => "/Users/test/.local/bin/iii",
localBinIiiPath: () => "/Users/test/.local/bin/iii",
iiiBinaryVersion: () => III_PINNED_VERSION,
+ externalEngineRunning: async () => false,
viewerReachable: async () => true,
runInit: async () => ({ ok: true, message: "wrote .env" }),
openEditor: async () => ({ ok: true, message: "saved" }),
@@ -338,11 +340,13 @@ describe("doctor v2 diagnostic catalog", () => {
expect(status.detail).toContain("ANTHROPIC_API_KEY");
});
- it("iii-on-path-not-local-bin warns when iii lives in another location", async () => {
+ it("iii-on-path-not-local-bin warns when iii lives in another location and the private install is missing", async () => {
+ const privatePath = "/Users/test/.agentmemory/bin/iii";
const diagnostics = buildDiagnostics(
stubEffects({
findIiiBinary: () => "/opt/homebrew/bin/iii",
- localBinIiiPath: () => "/Users/test/.local/bin/iii",
+ localBinIiiPath: () => privatePath,
+ iiiBinaryVersion: (bin) => (bin === privatePath ? null : III_PINNED_VERSION),
}),
);
const check = diagnostics.find((d) => d.id === "iii-on-path-not-local-bin")!;
@@ -351,6 +355,96 @@ describe("doctor v2 diagnostic catalog", () => {
expect(check.manualOnly).toBe(true);
});
+ it("both engine checks pass when the private install matches the pin despite a mismatched PATH iii", async () => {
+ const privatePath = "/Users/test/.agentmemory/bin/iii";
+ const diagnostics = buildDiagnostics(
+ stubEffects({
+ findIiiBinary: () => "/opt/homebrew/bin/iii",
+ localBinIiiPath: () => privatePath,
+ iiiBinaryVersion: (bin) => (bin === privatePath ? III_PINNED_VERSION : "0.99.99"),
+ }),
+ );
+ const versionCheck = diagnostics.find((d) => d.id === "engine-version-mismatch")!;
+ const versionStatus = await versionCheck.check(stubCtx());
+ expect(versionStatus.ok).toBe(true);
+ expect(versionStatus.detail).toContain("pinned private install is available for startup");
+ const pathCheck = diagnostics.find((d) => d.id === "iii-on-path-not-local-bin")!;
+ const pathStatus = await pathCheck.check(stubCtx());
+ expect(pathStatus.ok).toBe(true);
+ expect(pathStatus.detail).toContain("/opt/homebrew/bin/iii");
+ });
+
+ it("engine-version-mismatch fix converges once the installer writes the private binary", async () => {
+ const privatePath = "/Users/test/.agentmemory/bin/iii";
+ let privateVersion: string | null = null;
+ const diagnostics = buildDiagnostics(
+ stubEffects({
+ findIiiBinary: () => "/opt/homebrew/bin/iii",
+ localBinIiiPath: () => privatePath,
+ iiiBinaryVersion: (bin) => (bin === privatePath ? privateVersion : "0.99.99"),
+ runIiiInstaller: async () => {
+ privateVersion = III_PINNED_VERSION;
+ return { ok: true, message: "installed" };
+ },
+ }),
+ );
+ const check = diagnostics.find((d) => d.id === "engine-version-mismatch")!;
+ const before = await check.check(stubCtx());
+ expect(before.ok).toBe(false);
+ const fixResult = await check.fix(stubCtx());
+ expect(fixResult.ok).toBe(true);
+ const after = await check.check(stubCtx());
+ expect(after.ok).toBe(true);
+ });
+
+ it("does not infer an external engine version from a pinned private install", async () => {
+ const install = vi.fn(async () => ({ ok: true }));
+ const stop = vi.fn(async () => ({ ok: true }));
+ const start = vi.fn(async () => ({ ok: true }));
+ const diagnostics = buildDiagnostics(stubEffects({
+ externalEngineRunning: async () => true,
+ runIiiInstaller: install,
+ runStop: stop,
+ runStart: start,
+ }));
+ const check = diagnostics.find((d) => d.id === "engine-version-mismatch")!;
+ const status = await check.check(stubCtx());
+ expect(status.ok).toBe(false);
+ expect(status.detail).toBe("external engine version is unknown");
+ expect(canAutoFix(check, status)).toBe(false);
+ expect(dryRunPlan(stubCtx(), [{ diagnostic: check, status }]).join("\n"))
+ .toContain(`manual fix: ${status.manualFix}`);
+ expect((await check.fix(stubCtx())).ok).toBe(false);
+ expect(install).not.toHaveBeenCalled();
+ expect(stop).not.toHaveBeenCalled();
+ expect(start).not.toHaveBeenCalled();
+ });
+
+ it("canAutoFix is false for manualOnly diagnostics and true otherwise", () => {
+ const diagnostics = buildDiagnostics(stubEffects());
+ const manual = diagnostics.find((d) => d.id === "iii-on-path-not-local-bin")!;
+ expect(canAutoFix(manual)).toBe(false);
+ const auto = diagnostics.find((d) => d.id === "engine-version-mismatch")!;
+ expect(canAutoFix(auto)).toBe(true);
+ });
+
+ it("accepts the pinned private install when iii is absent from PATH", async () => {
+ const diagnostics = buildDiagnostics(stubEffects({ findIiiBinary: () => null }));
+ for (const id of ["engine-version-mismatch", "iii-on-path-not-local-bin"]) {
+ expect((await diagnostics.find((d) => d.id === id)!.check(stubCtx())).ok).toBe(true);
+ }
+ });
+
+ it("rejects a missing or mismatched private install when PATH has no iii", async () => {
+ for (const version of [null, "0.99.99"]) {
+ const diagnostics = buildDiagnostics(stubEffects({
+ findIiiBinary: () => null,
+ iiiBinaryVersion: () => version,
+ }));
+ expect((await diagnostics.find((d) => d.id === "engine-version-mismatch")!.check(stubCtx())).ok).toBe(false);
+ }
+ });
+
it("dryRunPlan lists each failing diagnostic with the fix preview", () => {
const diagnostics = buildDiagnostics(stubEffects());
const results = diagnostics.map((d) => ({
@@ -372,6 +466,14 @@ describe("doctor v2 diagnostic catalog", () => {
expect(lines.length).toBe(1);
expect(lines[0]).toContain("All checks passing");
});
+
+ it("dryRunPlan labels manual-only diagnostics without promising automatic fixes", () => {
+ const diagnostic = buildDiagnostics(stubEffects())
+ .find((d) => d.id === "iii-on-path-not-local-bin")!;
+ const plan = dryRunPlan(stubCtx(), [{ diagnostic, status: { ok: false } }]).join("\n");
+ expect(plan).toContain(`manual fix: ${diagnostic.fixPreview}`);
+ expect(plan).not.toContain("would fix:");
+ });
});
describe("parseEnvFile", () => {
diff --git a/test/website-meta.test.ts b/test/website-meta.test.ts
index 1bf61123..bb4fe033 100644
--- a/test/website-meta.test.ts
+++ b/test/website-meta.test.ts
@@ -3,15 +3,60 @@ import { execFileSync } from "node:child_process";
import { copyFileSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import { join } from "node:path";
+import { EVALUATION, evaluationProse } from "../website/lib/bench.js";
describe("website metadata", () => {
- it("counts REST routes regardless of config field order and includes viewer endpoints", () => {
+ it("updates benchmark prose for changed coverage, failures and strict scores", () => {
+ const data = structuredClone(EVALUATION);
+ Object.assign(data.longmemeval.byAdapter["agentmemory-bm25-component"], {
+ n: 10, completed: 9, failed: 1, scored: 7,
+ answerable: { attempted: 8, completed: 7, failed: 1 }, unanswerable: 2,
+ });
+ Object.assign(data.longmemeval.byAdapter.grep, {
+ n: 11, completed: 10, failed: 1, scored: 9,
+ answerable: { attempted: 9, completed: 9, failed: 0 }, unanswerable: 2,
+ });
+ Object.assign(data.longmemeval.byType["multi-session"]["agentmemory-bm25-component"], {
+ n: 5, scored: 3, recallAll: 2 / 3, hit: 3,
+ });
+ Object.assign(data.coding.byAdapter["agentmemory-bm25-component"], {
+ n: 4, completed: 3, failed: 1, scored: 3, recallAll: 2 / 3,
+ });
+ Object.assign(data.coding.byAdapter.grep, {
+ n: 4, completed: 4, failed: 0, scored: 4, recallAll: 1,
+ });
+
+ const summary = evaluationProse(data);
+ expect(summary.longmemeval).toContain("19 of 21 adapter attempts completed, with 2 execution failures");
+ expect(summary.longmemeval).toContain("BM25 component: 8 answerable and 2 unanswerable");
+ expect(summary.longmemeval).toContain("grep baseline: 9 answerable and 2 unanswerable");
+ expect(summary.multiSession).toContain("2/3 (66.7%)");
+ expect(summary.coding).toContain("BM25 component 2/3; grep baseline 4/4");
+ expect(summary.coding).toContain("7 of 8 adapter attempts completed, with 1 execution failure.");
+ });
+
+ it("derives source counts and refuses stale metadata or incomplete evaluation receipts", () => {
const root = mkdtempSync(join(tmpdir(), "agentmemory-website-meta-"));
try {
const scripts = join(root, "website", "scripts");
const triggers = join(root, "src", "triggers");
mkdirSync(scripts, { recursive: true });
mkdirSync(triggers, { recursive: true });
+ mkdirSync(join(root, "scripts"));
+ mkdirSync(join(root, "src", "mcp"));
+ writeFileSync(join(root, "scripts", "docs-sync.state.json"), JSON.stringify({
+ version: "1.2.3", facts: { tests: ["4000+"] },
+ }));
+ writeFileSync(join(root, "src", "mcp", "tools-registry.ts"), 'const tool = { name: "memory_save" };');
+ writeFileSync(join(root, "src", "types.ts"), 'export type HookType = "session-start" | "stop";');
+ const receipts = "docs/benchmarks/receipts/2026-10-28";
+ for (const suite of ["longmemeval", "coding"]) {
+ const target = join(root, receipts, suite);
+ mkdirSync(target, { recursive: true });
+ for (const name of ["manifest.json", "summary.json"]) {
+ copyFileSync(join(receipts, suite, name), join(target, name));
+ }
+ }
writeFileSync(join(root, "package.json"), JSON.stringify({ version: "1.2.3" }));
copyFileSync("website/scripts/gen-meta.mjs", join(scripts, "gen-meta.mjs"));
writeFileSync(join(root, "offline.mjs"), "globalThis.fetch = async () => { throw new Error('offline'); };\n");
@@ -33,6 +78,18 @@ describe("website metadata", () => {
const meta = JSON.parse(readFileSync(join(root, "website", "lib", "generated-meta.json"), "utf8"));
expect(meta.version).toBe("1.2.3");
expect(meta.restEndpoints).toBe(6);
+ expect(meta.mcpTools).toBe(1);
+ expect(meta.hooks).toBe(2);
+ expect(meta.testCases).toBe("4000+");
+ expect(meta).not.toHaveProperty("testsPassing");
+ const generated = JSON.parse(readFileSync(join(root, "website", "lib", "generated-evaluation.json"), "utf8"));
+ expect(generated.longmemeval).toEqual(JSON.parse(readFileSync(join(receipts, "longmemeval", "summary.json"), "utf8")));
+
+ const run = () => execFileSync(process.execPath, ["--import", join(root, "offline.mjs"), join(scripts, "gen-meta.mjs")], { timeout: 10_000, stdio: "pipe" });
+ writeFileSync(join(root, receipts, "longmemeval", "summary.json"), JSON.stringify({ planned: 1000, completed: 999, failed: 0 }));
+ expect(run).toThrow("incomplete longmemeval evaluation receipts");
+ writeFileSync(join(root, "package.json"), JSON.stringify({ version: "1.2.4" }));
+ expect(run).toThrow("source metadata is missing or stale");
} finally {
rmSync(root, { recursive: true, force: true });
}
diff --git a/website/app/benchmarks/page.tsx b/website/app/benchmarks/page.tsx
index 6070b234..55843955 100644
--- a/website/app/benchmarks/page.tsx
+++ b/website/app/benchmarks/page.tsx
@@ -1,243 +1,58 @@
import type { Metadata } from "next";
import { DocPage, Prose } from "@/components/site/Doc";
import { QuestionDots } from "@/components/site/BenchFigures";
-import { LONGMEMEVAL, TOKENS, src } from "@/lib/site";
-import { BUDGETS, RECOVERY_FACTS, REPRODUCE, TOKEN_SETUP, TYPE_HITS, benchSrc } from "@/lib/bench";
+import { REPO_URL, src } from "@/lib/site";
+import { ADAPTERS, ADAPTER_LABELS, EVALUATION, REPRODUCE, evaluationProse, percent } from "@/lib/bench";
import s from "./page.module.css";
export const metadata: Metadata = {
title: "Benchmarks",
- description:
- "How agentmemory is measured: LongMemEval-S retrieval recall, token cost per query, crash recovery and the resource budgets the cost bench enforces. Every number links to its source file.",
+ description: "Reproducible retrieval evaluation with strict recall-all, coverage, baselines and limitations.",
alternates: { canonical: "/benchmarks" },
};
-const pct = (n: number) => `${n.toFixed(1)}%`;
-const mib = (kib: number) => `${(kib / 1024).toFixed(0)} MiB`;
-const kib = (bytes: number) => `${(bytes / 1024).toFixed(1)} KiB`;
-const sec = (ms: number) => `${(ms / 1000).toFixed(1)} s`;
-
-const TOKEN_ROWS = [
- { name: "Load every observation", r5: 37.0, r10: 55.8, p5: 78.0, ndcg: 80.3, mrr: 82.5, tokens: TOKENS.fullContext },
- { name: "BM25 only, keyless", r5: 43.8, r10: 55.9, p5: 95.0, ndcg: 82.7, mrr: 95.5, tokens: TOKENS.agentmemory },
- { name: "BM25 + local vectors", r5: 43.8, r10: 64.1, p5: 98.0, ndcg: 94.9, mrr: 100.0, tokens: TOKENS.agentmemory },
-];
-
-function Src({ href, children }: { href: string; children: string }) {
- return (
- <a className="src" href={href} target="_blank" rel="noreferrer">
- {children}
- </a>
- );
-}
-
export default function BenchmarksPage() {
- return (
- <DocPage
- eyebrow="Benchmarks"
- title="Measured in the open, reproducible on your machine."
- lede="Retrieval quality, token cost, crash recovery and resource budgets. Each table names the file it comes from and the command that regenerates it."
- >
- <div className={s.figure}>
- <QuestionDots />
- </div>
-
- <Prose>
- <h2>What is measured</h2>
- <ul>
- <li>
- <strong>Dataset.</strong> LongMemEval-S: {LONGMEMEVAL.questions} questions, about 48 sessions and 115K tokens
- per question. <Src href={src.dataset}>dataset</Src> <Src href={src.paper}>paper</Src>
- </li>
- <li>
- <strong>Metric.</strong> recall_any@K: does any gold session appear in the top K results. Each question builds
- a fresh index from its own sessions and searches with the question text.
- </li>
- <li>
- <strong>Embeddings.</strong> {LONGMEMEVAL.embedding}. No API key and no LLM in the loop.
- </li>
- <li>
- <strong>What it is not.</strong> This is retrieval recall, not end-to-end answer accuracy. The official
- LongMemEval score adds answer generation and an LLM judge, and we do not claim one.
- </li>
- </ul>
-
- <h2>Retrieval recall</h2>
- <div className={s.tableWrap}>
- <table>
- <thead>
- <tr>
- <th>Mode</th>
- <th>R@5</th>
- <th>R@10</th>
- <th>R@20</th>
- <th>NDCG@10</th>
- <th>MRR</th>
- </tr>
- </thead>
- <tbody>
- {[
- { name: "BM25 + vectors", d: LONGMEMEVAL.hybrid },
- { name: "BM25 only, keyless", d: LONGMEMEVAL.bm25 },
- ].map((r) => (
- <tr key={r.name}>
- <td>{r.name}</td>
- <td>{pct(r.d.r5)}</td>
- <td>{pct(r.d.r10)}</td>
- <td>{pct(r.d.r20)}</td>
- <td>{pct(r.d.ndcg10)}</td>
- <td>{pct(r.d.mrr)}</td>
- </tr>
- ))}
- </tbody>
- </table>
- </div>
- <p>
- Adding local vectors to BM25 lifts R@5 from {pct(LONGMEMEVAL.bm25.r5)} to {pct(LONGMEMEVAL.hybrid.r5)}.{" "}
- <Src href={src.longmemeval}>benchmark/LONGMEMEVAL.md</Src>
- </p>
-
- <h3>By question type, BM25 + vectors</h3>
- <div className={s.tableWrap}>
- <table>
- <thead>
- <tr>
- <th>Type</th>
- <th>Questions</th>
- <th>R@5</th>
- <th>R@10</th>
- <th>Found at top 5</th>
- </tr>
- </thead>
- <tbody>
- {LONGMEMEVAL.byType.map((t) => (
- <tr key={t.type}>
- <td>{t.type}</td>
- <td>{t.count}</td>
- <td>{pct(t.r5)}</td>
- <td>{pct(t.r10)}</td>
- <td>
- {TYPE_HITS.find((h) => h.type === t.type)?.hits[5]} of {t.count}
- </td>
- </tr>
- ))}
- </tbody>
- </table>
- </div>
- <p>
- Preferences are the hardest type: they depend on implicit statements, and R@5 is {pct(LONGMEMEVAL.byType[5].r5)}.
- Knowledge updates and multi-session questions are the strongest.
- </p>
-
- <h2>Token cost per query</h2>
- <p>
- A smaller labeled set of coding-session observations: {TOKEN_SETUP.observations} observations across{" "}
- {TOKEN_SETUP.sessions} sessions and {TOKEN_SETUP.queries} queries, with {TOKEN_SETUP.model}. The baseline loads
- every observation into the prompt.
- </p>
- <div className={s.tableWrap}>
- <table>
- <thead>
- <tr>
- <th>Mode</th>
- <th>R@5</th>
- <th>R@10</th>
- <th>P@5</th>
- <th>NDCG@10</th>
- <th>MRR</th>
- <th>Tokens</th>
- </tr>
- </thead>
- <tbody>
- {TOKEN_ROWS.map((r) => (
- <tr key={r.name}>
- <td>{r.name}</td>
- <td>{pct(r.r5)}</td>
- <td>{pct(r.r10)}</td>
- <td>{pct(r.p5)}</td>
- <td>{pct(r.ndcg)}</td>
- <td>{pct(r.mrr)}</td>
- <td>{r.tokens.toLocaleString("en-US")}</td>
- </tr>
- ))}
- </tbody>
- </table>
- </div>
- <p>
- {TOKENS.agentmemory.toLocaleString("en-US")} tokens instead of {TOKENS.fullContext.toLocaleString("en-US")} is{" "}
- {TOKENS.savedPct}% fewer, with higher precision, measured on v0.6.0.{" "}
- <Src href={src.tokens}>benchmark/REAL-EMBEDDINGS.md</Src>
- </p>
-
- <h2>Crash recovery</h2>
- <p>
- The cost bench runs the built package in an isolated home, captures N observations, kills the whole process
- tree with SIGKILL, boots it again and checks that nothing was lost.
- </p>
- <ul className={s.facts}>
- {RECOVERY_FACTS.map((f) => (
- <li key={f.value}>
- <strong>{f.value}</strong> {f.label}
- </li>
- ))}
- </ul>
- <p>
- <Src href={benchSrc.costsFigure}>docs/benchmarks/capture-costs.svg</Src>{" "}
- <Src href={benchSrc.readme}>benchmark/README.md</Src>
- </p>
-
- <h2>Resource budgets</h2>
- <p>
- The cost bench checks each run against these ceilings. They were derived from repeated runs, taking the worst run plus
- headroom, so they are upper limits rather than typical values, and they depend on the machine that runs them.
- The embed profile uses 768-dimension vectors.
- </p>
- <div className={s.tableWrap}>
- <table>
- <thead>
- <tr>
- <th>Profile</th>
- <th>Observations</th>
- <th>Hook p95</th>
- <th>Index ready after kill</th>
- <th>Disk per observation</th>
- <th>Idle memory</th>
- <th>Search tokens p50</th>
- </tr>
- </thead>
- <tbody>
- {BUDGETS.map((b) => (
- <tr key={`${b.profile}-${b.n}`}>
- <td>{b.profile}</td>
- <td>{b.n.toLocaleString("en-US")}</td>
- <td>{b.hookP95Ms} ms</td>
- <td>{sec(b.recoveryIndexReadyMs)}</td>
- <td>{kib(b.diskPerObsBytes)}</td>
- <td>{mib(b.idleRssKiB)}</td>
- <td>{b.searchTokensP50}</td>
- </tr>
- ))}
- </tbody>
- </table>
- </div>
- <p>
- <Src href={benchSrc.budgets}>benchmark/capture-costs-budgets.json</Src>
- </p>
-
- <h2>Reproduce</h2>
- <p>Retrieval, from the repository root, after downloading the dataset:</p>
- <pre>
- <code>{`${REPRODUCE.download}\n\n${REPRODUCE.bm25}\n${REPRODUCE.hybrid}`}</code>
- </pre>
- <p>Costs and crash recovery, against the built package:</p>
- <pre>
- <code>{REPRODUCE.costs}</code>
- </pre>
- <p>
- Scripts: <Src href={benchSrc.longmemevalScript}>benchmark/longmemeval-bench.ts</Src>{" "}
- <Src href={benchSrc.realEmbeddingsScript}>benchmark/real-embeddings-eval.ts</Src>
- </p>
- </Prose>
- </DocPage>
- );
+ const summary = evaluationProse();
+ return <DocPage eyebrow="Benchmarks" title="Complete evidence, measured in the open."
+ lede="AgentMemory 0.9.31 development checkout, measured on 2026-10-28. Every result has a dataset fingerprint and per-question receipts.">
+ <div className={s.figure}><QuestionDots /></div>
+ <Prose>
+ <h2>What the scores mean</h2>
+ <p>Strict recall-all@5 requires every gold source session in the first five distinct results. Fractional recall measures how much of the required evidence was found. Any-hit needs just one source.</p>
+ <p>Each question builds a fresh index over its supplied sessions. This run tests the production BM25 SearchIndex and a token-overlap grep baseline. It does not run the daemon, capture hooks, compression, vectors, graph or an answer-generating model.</p>
+ <p>{summary.longmemeval} Unanswerable questions count toward coverage, but have no retrieval score; abstention quality is not measured.</p>
+ <p>{summary.multiSession} Finding one relevant session can still leave an answer without all its evidence.</p>
+ <p><a href={src.longmemeval}>Full report, per-type results, manifests and compressed per-question receipts</a>.</p>
+
+ <h2>Coding-session regression fixture</h2>
+ <div className={s.tableWrap}><table>
+ <thead><tr><th>Adapter</th><th>Strict recall-all@5</th><th>Completed</th><th>Failures</th></tr></thead>
+ <tbody>{ADAPTERS.map((adapter) => {
+ const r = EVALUATION.coding.byAdapter[adapter];
+ return <tr key={adapter}><td>{ADAPTER_LABELS[adapter]}</td><td>{percent(r.recallAll)}</td><td>{r.completed}/{r.n}</td><td>{r.failed}</td></tr>;
+ })}</tbody>
+ </table></div>
+ <p>{summary.coding} This small fixture is useful regression coverage, not evidence that a system wins across coding workloads.</p>
+
+ <h2>Reproduce</h2>
+ <p>From the repository root with dependencies installed, download the pinned public dataset:</p>
+ <pre><code>{REPRODUCE.download}</code></pre>
+ <p>Expected SHA256: <code>d6f21ea9d60a0d56f34a05b609c79c88a451d2ae03597821ea3d5a9678c3a442</code>.</p>
+ <pre><code>{`${REPRODUCE.retrieval}\n${REPRODUCE.coding}`}</code></pre>
+ <p>The loader strips answer labels, uses opaque IDs, retains dates and coalesces 13 identical repeated distractor IDs. See the <a href={src.evaluation}>evaluation guide</a> for the complete input and scoring contract.</p>
+
+ <h2>Release improvements and runtime checks</h2>
+ <p>0.9.31 adds scoped saved-memory recall, incremental summaries, paged exports, validated imports and native Pi, OMP and DeepSeek Harness integrations, alongside compression and recovery fixes. <a href={`${src.releases}/tag/v0.9.31`}>Release notes</a>.</p>
+ <p>Lifecycle and packaged startup/restart behavior are checked separately by the <a href={src.evaluation}>live evaluation suites</a> and <a href={src.gate}>release gate</a>. The component results above do not certify those paths or establish a speedup over the previous release.</p>
+
+ <h2>Historical measurements</h2>
+ <ul>
+ <li><a href={`${REPO_URL}/blob/main/benchmark/LONGMEMEVAL.md`}>Hybrid LongMemEval experiment</a>: older any-hit scoring and input pipeline, not comparable to strict recall-all.</li>
+ <li><a href={src.tokens}>v0.6.0 token experiment</a>: fixture-specific context savings, not a per-session guarantee.</li>
+ <li><a href={src.agentLife}>May coding-session benchmark</a>: older full-runtime experiment.</li>
+ <li><a href={`${REPO_URL}/blob/main/benchmark/capture-costs-budgets.json`}>Capture-cost budgets</a>: machine-dependent regression ceilings, not typical latency or a current speed claim.</li>
+ </ul>
+ <p>We have not run a matched evaluation of the other memory products. Compare products by <a href="/vs">deployment and use case</a>; answer accuracy, retrieval recall and token use require different measurements.</p>
+ </Prose>
+ </DocPage>;
}
diff --git a/website/app/opengraph-image.tsx b/website/app/opengraph-image.tsx
index d962edf9..cff70b22 100644
--- a/website/app/opengraph-image.tsx
+++ b/website/app/opengraph-image.tsx
@@ -1,6 +1,5 @@
import { ImageResponse } from "next/og";
import { compact, getProjectMeta } from "@/lib/meta";
-import { LONGMEMEVAL } from "@/lib/site";
export const alt = "agentmemory: memory that outlives the session";
export const size = { width: 1200, height: 630 };
@@ -103,8 +102,8 @@ export default function Image() {
}}
>
<div style={{ display: "flex", gap: 10 }}>
- <span style={{ color: INK }}>{LONGMEMEVAL.hybrid.r5}%</span>
- <span>recall@5 LongMemEval-S</span>
+ <span style={{ color: INK }}>{meta.mcpTools}</span>
+ <span>MCP tools</span>
</div>
<div style={{ display: "flex", gap: 10 }}>
<span style={{ color: INK }}>{compact(meta.stars)}</span>
diff --git a/website/app/page.tsx b/website/app/page.tsx
index 1d69aafe..a321d6d6 100644
--- a/website/app/page.tsx
+++ b/website/app/page.tsx
@@ -25,6 +25,7 @@ export default function Page() {
<Hero />
<Proof />
<Featured />
+ <Interfaces />
<UseCases />
<ContextFunnel />
<InfraTiles />
@@ -34,7 +35,6 @@ export default function Page() {
<Benchmarks />
<Agents />
<LocalFirst />
- <Interfaces />
<Faq />
<Install />
</main>
diff --git a/website/app/privacy/page.tsx b/website/app/privacy/page.tsx
index 78b0d076..8fb570d4 100644
--- a/website/app/privacy/page.tsx
+++ b/website/app/privacy/page.tsx
@@ -16,13 +16,14 @@ export default function PrivacyPage() {
>
<Prose>
<p>
- <strong>What it stores.</strong> Memories, and the agent activity you choose to capture, are saved in a data folder on your machine, for example{" "}
+ <strong>What it stores.</strong> Memories, and the agent activity you choose to capture, are saved in a data folder on your machine by default, for example{" "}
<code>~/Library/Application Support/agentmemory</code> on macOS. You can export or delete them at any time.
</p>
<p>
- <strong>What leaves your machine.</strong> Nothing, unless you configure it. If you add an LLM or embedding
- provider, agentmemory sends that provider the text it needs, under the provider&apos;s own privacy policy. Local
- embeddings send nothing.
+ <strong>What leaves your machine.</strong> Memory text stays local with local storage and keyless search or local models.
+ Keyless search does not guarantee local storage: a configured remote Redis backend stores memory records remotely.
+ Configured remote services receive the text they process, under their own privacy policies.
+ Installation, updates and the first local model download require network access.
</p>
<p>
<strong>Telemetry.</strong> None.
diff --git a/website/app/vs/[slug]/page.tsx b/website/app/vs/[slug]/page.tsx
index 7e11eafd..15780f13 100644
--- a/website/app/vs/[slug]/page.tsx
+++ b/website/app/vs/[slug]/page.tsx
@@ -72,8 +72,7 @@ export default async function Page({ params }: { params: Promise<{ slug: string
</table>
</div>
<p className={s.note}>
- Benchmark figures are each project&apos;s own published results. Different benchmarks, metrics and harnesses are
- not directly comparable.
+ No competitor was run in our current evaluation. Product capabilities are sourced separately; compare scores only with matched datasets, metrics and configurations.
</p>
<h2>How they differ</h2>
diff --git a/website/components/site/Agents.module.css b/website/components/site/Agents.module.css
index 30d52c3e..057415ba 100644
--- a/website/components/site/Agents.module.css
+++ b/website/components/site/Agents.module.css
@@ -156,10 +156,6 @@
overflow-wrap: anywhere;
}
-.slug {
- white-space: nowrap;
-}
-
.cmd :global(.copy-state) {
flex: none;
color: var(--muted);
diff --git a/website/components/site/Agents.tsx b/website/components/site/Agents.tsx
index 0866ceba..9e11d20c 100644
--- a/website/components/site/Agents.tsx
+++ b/website/components/site/Agents.tsx
@@ -29,9 +29,7 @@ function Card({ agent, index }: { agent: Agent; index: number }) {
{agent.detail !== LABEL[agent.support] ? <span className={s.detail}>{agent.detail}</span> : <span />}
{cmd ? (
<CopyButton text={cmd} label={`Copy ${cmd}`} className={s.cmd}>
- <code>
- agentmemory connect <span className={s.slug}>{agent.connect}</span>
- </code>
+ <code>{cmd}</code>
</CopyButton>
) : (
<span className={`mono ${s.cmdNone}`}>add the MCP block below</span>
@@ -46,7 +44,7 @@ export function Agents() {
const plugins = AGENTS.filter((a) => a.support === "plugin");
const rest = AGENTS.filter((a) => a.support !== "plugin");
const groups = [
- { title: "Native plugins", note: "Hooks capture automatically. Nothing to remember to call.", items: plugins, offset: 0 },
+ { title: "Native plugins", note: "Capture requires an installed native plugin or hooks. Use --with-hooks where hook capture is opt-in.", items: plugins, offset: 0 },
{ title: "MCP and REST", note: "Any MCP client gets the memory tools. Anything that speaks HTTP can use the REST API.", items: rest, offset: plugins.length },
];
return (
@@ -58,8 +56,7 @@ export function Agents() {
{AGENTS.length} agents. One memory.
</h2>
<p>
- Switch agents mid-project and the memory comes with you. Every agent below reads and writes the same store on
- your machine.
+ Switch agents mid-project and the memory comes with you. Every agent below reads and writes the same memory store.
</p>
</div>
{groups.map((g) => (
diff --git a/website/components/site/BenchFigures.tsx b/website/components/site/BenchFigures.tsx
index 80a1a71e..16447742 100644
--- a/website/components/site/BenchFigures.tsx
+++ b/website/components/site/BenchFigures.tsx
@@ -1,8 +1,7 @@
"use client";
import { useEffect, useRef, useState, type CSSProperties } from "react";
-import { LONGMEMEVAL } from "@/lib/site";
-import { TYPE_HITS, totalHits, type K } from "@/lib/bench";
+import { ADAPTERS, ADAPTER_LABELS, EVALUATION, TYPE_HITS, totalHits, percent, type Adapter } from "@/lib/bench";
import s from "./Benchmarks.module.css";
function useInView<T extends HTMLElement>() {
@@ -30,14 +29,6 @@ function useInView<T extends HTMLElement>() {
return [ref, inView] as const;
}
-const LEDGER: { key: keyof typeof LONGMEMEVAL.hybrid; label: string; k?: K }[] = [
- { key: "r5", label: "R@5", k: 5 },
- { key: "r10", label: "R@10", k: 10 },
- { key: "r20", label: "R@20" },
- { key: "ndcg10", label: "NDCG@10" },
- { key: "mrr", label: "MRR" },
-];
-
function dotDelay(i: number, lit: boolean, before: number, firstReveal: boolean): number {
if (firstReveal) return Math.min(i * 4, 420);
if (lit && i >= before) return Math.min((i - before) * 45, 400);
@@ -46,24 +37,24 @@ function dotDelay(i: number, lit: boolean, before: number, firstReveal: boolean)
export function QuestionDots() {
const [ref, inView] = useInView<HTMLDivElement>();
- const [k, setK] = useState<K>(5);
- const [prevK, setPrevK] = useState<K | null>(null);
- const hits = totalHits(k);
+ const [adapter, setAdapter] = useState<Adapter>("agentmemory-bm25-component");
+ const [previous, setPrevious] = useState<Adapter | null>(null);
+ const hits = totalHits(adapter);
- const choose = (next: K) => {
- if (next === k) return;
- setPrevK(k);
- setK(next);
+ const choose = (next: Adapter) => {
+ if (next === adapter) return;
+ setPrevious(adapter);
+ setAdapter(next);
};
return (
<div ref={ref} className={s.dotsFigure}>
<div className={s.figHead}>
- <span className="mono">LongMemEval-S · 500 questions · one dot each</span>
- <div className={s.toggle} role="group" aria-label="Retrieved sessions per question">
- {([5, 10] as K[]).map((v) => (
- <button key={v} type="button" aria-pressed={k === v} onClick={() => choose(v)}>
- top {v}
+ <span className="mono">LongMemEval-S · {EVALUATION.longmemeval.byAdapter[adapter].scored} answerable questions</span>
+ <div className={s.toggle} role="group" aria-label="Retrieval adapter">
+ {ADAPTERS.map((v) => (
+ <button key={v} type="button" aria-pressed={adapter === v} onClick={() => choose(v)}>
+ {ADAPTER_LABELS[v]}
</button>
))}
</div>
@@ -71,15 +62,15 @@ export function QuestionDots() {
<p className={s.verdict}>
<strong>
- {hits} of {LONGMEMEVAL.questions}
+ {hits} of {EVALUATION.longmemeval.byAdapter[adapter].scored}
</strong>{" "}
- questions found a gold session in the top {k} results.
+ questions found every required source in the top 5 results.
</p>
<div className={s.groups}>
{TYPE_HITS.map((t) => {
- const on = t.hits[k];
- const before = prevK ? t.hits[prevK] : 0;
+ const on = t.hits[adapter];
+ const before = previous ? t.hits[previous] : 0;
return (
<div key={t.type} className={s.group}>
<div className={s.groupHead}>
@@ -91,7 +82,7 @@ export function QuestionDots() {
<div className={s.dots} aria-hidden="true">
{Array.from({ length: t.count }, (_, i) => {
const lit = inView && i < on;
- const delay = dotDelay(i, lit, before, prevK === null);
+ const delay = dotDelay(i, lit, before, previous === null);
return (
<span key={i} className={s.dot} data-on={lit || undefined} data-hit={i < on || undefined}>
<span className={s.fill} style={{ "--d": `${delay}ms` } as CSSProperties} />
@@ -106,10 +97,10 @@ export function QuestionDots() {
<div className={s.legend}>
<span className="mono">
- <span className={`${s.key} ${s.keyOn}`} /> gold session retrieved
+ <span className={`${s.key} ${s.keyOn}`} /> all required sources retrieved
</span>
<span className="mono">
- <span className={s.key} /> missed
+ <span className={s.key} /> incomplete
</span>
</div>
@@ -117,31 +108,22 @@ export function QuestionDots() {
<table className={s.ledger}>
<thead>
<tr>
- <th scope="col">recall mode</th>
- {LEDGER.map((c) => (
- <th key={c.key} scope="col" data-active={c.k === k || undefined}>
- {c.label}
- </th>
- ))}
+ <th scope="col">Adapter</th>
+ <th scope="col">Strict recall-all@5</th>
+ <th scope="col">Fractional recall@5</th>
+ <th scope="col">Any-hit@5</th>
</tr>
</thead>
<tbody>
- <tr>
- <th scope="row">BM25 + vectors</th>
- {LEDGER.map((c) => (
- <td key={c.key} data-active={c.k === k || undefined}>
- {LONGMEMEVAL.hybrid[c.key].toFixed(1)}%
- </td>
- ))}
- </tr>
- <tr>
- <th scope="row">BM25 only, keyless</th>
- {LEDGER.map((c) => (
- <td key={c.key} data-active={c.k === k || undefined}>
- {LONGMEMEVAL.bm25[c.key].toFixed(1)}%
- </td>
- ))}
- </tr>
+ {ADAPTERS.map((key) => {
+ const result = EVALUATION.longmemeval.byAdapter[key];
+ return <tr key={key}>
+ <th scope="row">{ADAPTER_LABELS[key]}</th>
+ <td>{percent(result.recallAll)}</td>
+ <td>{percent(result.r)}</td>
+ <td>{percent(result.hit / result.scored)}</td>
+ </tr>;
+ })}
</tbody>
</table>
</div>
diff --git a/website/components/site/Benchmarks.tsx b/website/components/site/Benchmarks.tsx
index ba1a5bd1..bd2f3cef 100644
--- a/website/components/site/Benchmarks.tsx
+++ b/website/components/site/Benchmarks.tsx
@@ -1,20 +1,24 @@
import Link from "next/link";
-import { LONGMEMEVAL, src } from "@/lib/site";
-import { REPRODUCE } from "@/lib/bench";
+import { src } from "@/lib/site";
+import { EVALUATION, REPRODUCE, percent } from "@/lib/bench";
import { CopyButton, Reveal } from "./client";
import { QuestionDots } from "./BenchFigures";
import s from "./Benchmarks.module.css";
export function Benchmarks() {
+ const bm25 = EVALUATION.longmemeval.byAdapter["agentmemory-bm25-component"];
+ const baseline = EVALUATION.longmemeval.byAdapter.grep;
+ const improvement = ((bm25.recallAll - baseline.recallAll) * 100).toFixed(1);
return (
<section id="benchmarks" className="section" aria-labelledby="benchmarks-title">
<div className="wrap">
<div className="section-head">
<span className="eyebrow">Benchmarks</span>
- <h2 id="benchmarks-title">It finds the right session, and sends less.</h2>
+ <h2 id="benchmarks-title">Useful recall. No model API required.</h2>
<p>
- Retrieval measured on LongMemEval-S, an academic benchmark for long-term memory, with a local embedding
- model and no API key. Every number below links to the file it came from.
+ Keyless BM25 finds every required source for {percent(bm25.recallAll)} of answerable LongMemEval-S questions,
+ {" "}{improvement} percentage points above the lexical baseline on the same inputs.
+ At least one relevant source is found for {percent(bm25.hit / bm25.scored)}. No embedding model or paid API.
</p>
</div>
@@ -23,10 +27,11 @@ export function Benchmarks() {
</Reveal>
<p className={s.cite}>
<a className="src" href={src.longmemeval} target="_blank" rel="noreferrer">
- benchmark/LONGMEMEVAL.md
+ Results and per-question receipts
</a>
<span>
- {LONGMEMEVAL.questions} questions, about 48 sessions each. Retrieval recall, not end-to-end answer accuracy.
+ {bm25.n} questions per adapter, {EVALUATION.longmemeval.failed} execution failures.
+ {" "}{bm25.unanswerable} unanswerable questions excluded from retrieval metrics.
</span>
</p>
@@ -38,8 +43,7 @@ export function Benchmarks() {
<a className="src" href={src.dataset} target="_blank" rel="noreferrer">
Hugging Face
</a>
- , then run the same script we did. Embeddings use {LONGMEMEVAL.embedding}. The benchmark is described in
- the{" "}
+ , then run the same evaluation. The pinned dataset and checksum are in the report. See the{" "}
<a className="src" href={src.paper} target="_blank" rel="noreferrer">
LongMemEval paper
</a>
@@ -47,20 +51,20 @@ export function Benchmarks() {
</p>
</div>
<div className={s.cmds}>
- <CopyButton text={REPRODUCE.hybrid} label="Copy the hybrid benchmark command" className={s.cmd}>
+ <CopyButton text={REPRODUCE.retrieval} label="Copy the LongMemEval evaluation command" className={s.cmd}>
<span className={s.prompt} aria-hidden="true">
$
</span>
- <code>{REPRODUCE.hybrid}</code>
+ <code>{REPRODUCE.retrieval}</code>
</CopyButton>
- <CopyButton text={REPRODUCE.bm25} label="Copy the BM25-only benchmark command" className={s.cmd}>
+ <CopyButton text={REPRODUCE.coding} label="Copy the coding fixture evaluation command" className={s.cmd}>
<span className={s.prompt} aria-hidden="true">
$
</span>
- <code>{REPRODUCE.bm25}</code>
+ <code>{REPRODUCE.coding}</code>
</CopyButton>
<Link className="btn btn-ghost" href="/benchmarks">
- Full methodology and cost budgets
+ Methodology, limitations and historical results
</Link>
</div>
</div>
diff --git a/website/components/site/ContextFunnel.tsx b/website/components/site/ContextFunnel.tsx
index 7485d57f..fedc555b 100644
--- a/website/components/site/ContextFunnel.tsx
+++ b/website/components/site/ContextFunnel.tsx
@@ -1,7 +1,7 @@
"use client";
import { useEffect, useRef, type CSSProperties } from "react";
-import { LONGMEMEVAL, TOKENS, src } from "@/lib/site";
+import { src } from "@/lib/site";
import s from "./ContextFunnel.module.css";
const COLS = 15;
@@ -12,7 +12,7 @@ const AGENTS = [
];
const TOTAL = COLS * AGENTS.reduce((n, a) => n + a.rows, 0);
const SELECTED = new Set([3, 17, 29, 41, 52, 66, 79, 88, 101, 114, 126, 139, 151, 163, 176, 188]);
-const STEPS = ["capture", "dedupe", "rank", "compress"];
+const STEPS = ["capture", "dedupe", "rank", "budget"];
function tone(index: number) {
const row = Math.floor(index / COLS);
@@ -51,18 +51,16 @@ export function ContextFunnel() {
};
}, []);
- const saved = Math.round((1 - TOKENS.agentmemory / TOKENS.fullContext) * 100);
return (
<section id="value" className="section" aria-labelledby="value-h">
<div className="wrap">
<div className="section-head">
<span className="eyebrow">What your agent gets</span>
- <h2 id="value-h">The {100 - saved}% of your history that matters, every session.</h2>
+ <h2 id="value-h">Relevant context, within a budget you control.</h2>
<p>
- A month of work across agents is {TOKENS.fullContext.toLocaleString("en-US")} tokens. Pasting all of it buries
- the task. agentmemory hands the next session the {TOKENS.agentmemory.toLocaleString("en-US")} tokens that
- answer it, and leaves the rest of the window for the work.
+ Recall by project and agent, select saved memories, and expand a source when you need more detail.
+ The diagram illustrates selection from shared history; the amount returned depends on the query and budget.
</p>
</div>
@@ -70,9 +68,9 @@ export function ContextFunnel() {
<div className={`${s.col} ${s.history}`}>
<div className={s.colHead}>
<span className="mono">your history</span>
- <span className={`mono ${s.count}`}>{TOKENS.fullContext.toLocaleString("en-US")} tokens</span>
+ <span className={`mono ${s.count}`}>shared history</span>
</div>
- <div className={s.grid} role="img" aria-label={`${TOTAL} squares, about 100 tokens each, from three agents. ${SELECTED.size} of them are selected as relevant.`}>
+ <div className={s.grid} role="img" aria-label={`${TOTAL} illustrative history items from three agents. ${SELECTED.size} of them are selected as relevant.`}>
{Array.from({ length: TOTAL }, (_, i) => (
<span
key={i}
@@ -89,7 +87,7 @@ export function ContextFunnel() {
</li>
))}
<li className={s.legendNote}>
- <span className="mono">1 square ≈ 100 tokens</span>
+ <span className="mono">illustrative history items</span>
</li>
</ul>
</div>
@@ -124,7 +122,7 @@ export function ContextFunnel() {
<div className={s.ctx}>
<div className={s.injected}>
<span className={`mono ${s.injHead}`}>
- SessionStart · {TOKENS.agentmemory.toLocaleString("en-US")} tokens
+ relevant context
</span>
<div className={s.slots}>
{Array.from({ length: SELECTED.size }, (_, i) => (
@@ -142,24 +140,24 @@ export function ContextFunnel() {
<ul className={s.outcomes}>
<li>
- <span className={s.big}>{LONGMEMEVAL.hybrid.r5}%</span>
- <span className={s.what}>of questions find the right past session in the top 5</span>
- <a className="src" href={src.longmemeval} target="_blank" rel="noreferrer">
- LongMemEval-S
+ <span className={s.big}>Scoped</span>
+ <span className={s.what}>filter recall by project and agent</span>
+ <a className="src" href={src.recall} target="_blank" rel="noreferrer">
+ recall guide
</a>
</li>
<li>
- <span className={s.big}>{TOKENS.savedPct}%</span>
- <span className={s.what}>fewer input tokens than pasting your whole history</span>
- <a className="src" href={src.tokens} target="_blank" rel="noreferrer">
- 240 observations, 30 sessions, v0.6.0
+ <span className={s.big}>Budgeted</span>
+ <span className={s.what}>choose how much context to retrieve</span>
+ <a className="src" href={src.recall} target="_blank" rel="noreferrer">
+ context budgets
</a>
</li>
<li>
- <span className={s.big}>14 ms</span>
- <span className={s.what}>median recall, with nothing sent over the network</span>
- <a className="src" href={src.agentLife} target="_blank" rel="noreferrer">
- coding-agent-life-v1 on v0.9.26
+ <span className={s.big}>Expandable</span>
+ <span className={s.what}>follow a result back to its full source</span>
+ <a className="src" href={src.recall} target="_blank" rel="noreferrer">
+ source expansion
</a>
</li>
</ul>
diff --git a/website/components/site/Hero.tsx b/website/components/site/Hero.tsx
index de261762..3aeeb13c 100644
--- a/website/components/site/Hero.tsx
+++ b/website/components/site/Hero.tsx
@@ -1,4 +1,4 @@
-import { AGENT_INSTALL_PROMPT, INSTALL_CMD, LONGMEMEVAL, REPO_URL, src } from "@/lib/site";
+import { AGENT_INSTALL_PROMPT, INSTALL_CMD, REPO_URL, src } from "@/lib/site";
import { compact, getLiveProjectMeta } from "@/lib/meta";
import { CopyButton } from "./client";
import { MemoryField } from "./MemoryField";
@@ -22,9 +22,9 @@ export async function Hero() {
<span className={s.sep} />
<span>Open source, Apache-2.0</span>
</a>
- <HeroTitle rankHref="https://trendshift.io/repositories/25123" />
+ <HeroTitle />
<p className={s.lede}>
- Your coding agent starts every session from zero. agentmemory captures what it did, keeps it on your machine,
+ Keep project decisions across sessions and agents. agentmemory captures supported activity, stores it on your machine,
and hands the right context back next time. No account, no API key, no cloud.
</p>
@@ -71,18 +71,18 @@ export async function Hero() {
<dl className={s.facts}>
<div>
- <dt className="mono">LongMemEval-S recall@5</dt>
+ <dt className="mono">MCP tools</dt>
<dd>
- {LONGMEMEVAL.hybrid.r5}%{" "}
- <a className="src" href={src.longmemeval} target="_blank" rel="noreferrer">
+ {meta.mcpTools}{" "}
+ <a className="src" href={`${REPO_URL}#mcp-server`} target="_blank" rel="noreferrer">
source
</a>
</dd>
</div>
<div>
- <dt className="mono">Leaves your machine</dt>
+ <dt className="mono">Memory storage</dt>
<dd>
- Nothing, by default{" "}
+ Local by default{" "}
<a className="src" href="/security">
how
</a>
diff --git a/website/components/site/HeroTitle.module.css b/website/components/site/HeroTitle.module.css
index 85b32086..dc76a29a 100644
--- a/website/components/site/HeroTitle.module.css
+++ b/website/components/site/HeroTitle.module.css
@@ -12,11 +12,6 @@
white-space: nowrap;
}
-.rank {
- color: var(--accent-ink);
- text-decoration: none;
-}
-
.slot {
display: inline-grid;
vertical-align: bottom;
diff --git a/website/components/site/HeroTitle.tsx b/website/components/site/HeroTitle.tsx
index 2c11b003..6f17afaa 100644
--- a/website/components/site/HeroTitle.tsx
+++ b/website/components/site/HeroTitle.tsx
@@ -3,7 +3,7 @@
import { useEffect, useRef, useState } from "react";
import s from "./HeroTitle.module.css";
-const MODES = ["persistent", "cross-agent", "infinite", "shared", "local", "long-term"];
+const MODES = ["persistent", "cross-agent", "searchable", "shared", "local", "long-term"];
const AGENTS = [
{ name: "Claude Code", logo: "https://svgl.app/library/claude-ai-icon.svg" },
@@ -16,7 +16,7 @@ const AGENTS = [
const STEP = 2400;
-export function HeroTitle({ rankHref }: { rankHref: string }) {
+export function HeroTitle() {
const [mode, setMode] = useState(0);
const [agent, setAgent] = useState(0);
const ref = useRef<HTMLHeadingElement | null>(null);
@@ -44,11 +44,8 @@ export function HeroTitle({ rankHref }: { rankHref: string }) {
return (
<h1 ref={ref} className={s.title}>
- <span className="sr-only">The number one persistent memory for coding agents.</span>
+ <span className="sr-only">Persistent memory for coding agents.</span>
<span aria-hidden="true" className={s.line}>
- <a className={s.rank} href={rankHref} target="_blank" rel="noreferrer" tabIndex={-1}>
- #1
- </a>{" "}
<span className={s.slot}>
{MODES.map((m, i) => (
<span key={m} className={s.item} data-on={i === mode || undefined}>
diff --git a/website/components/site/InfraTiles.tsx b/website/components/site/InfraTiles.tsx
index ec75949b..eb337b0c 100644
--- a/website/components/site/InfraTiles.tsx
+++ b/website/components/site/InfraTiles.tsx
@@ -1,5 +1,5 @@
import type { CSSProperties } from "react";
-import { src } from "@/lib/site";
+
import { Reveal } from "./client";
import s from "./InfraTiles.module.css";
@@ -30,13 +30,13 @@ function Gauge() {
})}
<path d={arc(0, 0.28)} className={s.fill} pathLength={1} />
<text x="150" y="128" textAnchor="middle" className={s.readout}>
- 14 ms
+ Local
</text>
<text x="30" y="166" className={s.scale}>
- 0
+ BM25
</text>
<text x="270" y="166" className={s.scale} textAnchor="end">
- 50 ms
+ vectors
</text>
</svg>
);
@@ -111,21 +111,21 @@ export function InfraTiles() {
<section className={s.section} aria-label="How agentmemory runs">
<div className={`wrap ${s.row}`}>
<Reveal className={s.tile}>
- <h3 className="mono">Recall in 14 ms</h3>
- <p>Median hybrid search over BM25, local vectors and the graph. No network round trip.</p>
+ <h3 className="mono">Keyless recall</h3>
+ <p>Keyword search works without an API key. Enable local embeddings to add semantic retrieval.</p>
<div className={s.art}>
<Gauge />
</div>
<code className={s.call}>
memory_smart_search(&quot;auth refresh&quot;)<span className={s.caret} />
</code>
- <a className="src" href={src.agentLife} target="_blank" rel="noreferrer">
- p50, coding-agent-life-v1 on v0.9.26
+ <a className="src" href="/benchmarks" target="_blank" rel="noreferrer">
+ retrieval evaluation
</a>
</Reveal>
<Reveal className={s.tile} delay={80}>
<h3 className="mono">Isolated on your machine</h3>
- <p>Binds to 127.0.0.1 with auth on by default. Nothing leaves unless you add a provider.</p>
+ <p>Binds to 127.0.0.1 with auth on by default. Memory text stays local unless you configure a remote service.</p>
<div className={s.art}>
<Isometric />
</div>
@@ -135,7 +135,7 @@ export function InfraTiles() {
</Reveal>
<Reveal className={s.tile} delay={160}>
<h3 className="mono">Survives compaction</h3>
- <p>When your agent compacts its context, the PreCompact hook puts memory back before the next turn.</p>
+ <p>On supported integrations, enable the PreCompact hook to restore memory context around compaction.</p>
<div className={s.art}>
<Compaction />
</div>
diff --git a/website/components/site/Install.tsx b/website/components/site/Install.tsx
index 4e0fdba5..88d514eb 100644
--- a/website/components/site/Install.tsx
+++ b/website/components/site/Install.tsx
@@ -47,8 +47,8 @@ export function Install() {
<div className={s.stepBody}>
<h3>Wire an agent</h3>
<p>
- Writes the MCP entry for the agent you pick. Add <code>--with-hooks</code> for automatic capture where the
- agent supports it.
+ Connects the selected agent through its native integration or MCP. Add <code>--with-hooks</code> for
+ optional hook capture. Pi and OMP use <code>--with-mcp</code> to add the full toolset alongside their native tools.
</p>
<div className={s.picker} role="radiogroup" aria-label="Agent">
{CONNECTABLE.map((a) => (
diff --git a/website/components/site/Interfaces.module.css b/website/components/site/Interfaces.module.css
index 03999c4e..24afb51a 100644
--- a/website/components/site/Interfaces.module.css
+++ b/website/components/site/Interfaces.module.css
@@ -1,195 +1,135 @@
-.tabs {
- display: flex;
- gap: 0;
- border: 1px solid var(--line);
- border-radius: var(--radius-lg);
- padding: 4px;
- background: var(--surface);
- width: fit-content;
- max-width: 100%;
- overflow-x: auto;
- scrollbar-width: none;
- margin-bottom: var(--s-6);
-}
-
-.tabs::-webkit-scrollbar {
- display: none;
+.section {
+ padding: clamp(48px, 6vw, 80px) 0;
}
-.tab {
+.heading {
display: grid;
- gap: 2px;
- text-align: left;
- padding: 8px 16px;
- border-radius: var(--radius);
- font-size: var(--t-14);
- color: var(--muted);
- white-space: nowrap;
- transition: background-color var(--dur-1) ease, color var(--dur-1) ease, transform var(--dur-1) var(--ease-out);
-}
-
-.tab :global(.mono) {
- font-size: var(--t-12);
- color: var(--faint);
+ grid-template-columns: minmax(0, 1.3fr) minmax(0, 0.9fr);
+ align-items: end;
+ gap: var(--s-7);
+ margin-bottom: var(--s-7);
}
-.tab:hover {
- color: var(--ink);
-}
-
-.tab:active {
- transform: scale(0.97);
-}
-
-.tab[aria-selected="true"] {
- background: var(--ink);
- color: var(--bg);
-}
-
-.tab[aria-selected="true"] :global(.mono) {
- color: color-mix(in srgb, var(--bg) 65%, transparent);
-}
-
-.body {
- display: grid;
- grid-template-columns: minmax(0, 0.8fr) minmax(0, 1.6fr);
- gap: clamp(24px, 4vw, 56px);
- align-items: start;
+.heading h2 {
+ margin-top: var(--s-4);
+ font-size: var(--t-49);
+ letter-spacing: var(--track-display);
+ line-height: 1.08;
}
-.text {
- display: grid;
- gap: var(--s-4);
- animation: in var(--dur-3) var(--ease-out) both;
-}
-
-.text h3 {
- font-size: var(--t-25);
-}
-
-.text p {
+.heading p {
color: var(--muted);
font-size: var(--t-16);
+ max-width: 400px;
}
-.text ul {
- list-style: none;
+.tour {
margin: 0;
- padding: 0;
- display: grid;
- gap: 8px;
- font-size: var(--t-14);
- color: var(--ink-2);
-}
-
-.text li {
- display: flex;
- gap: 10px;
- align-items: baseline;
-}
-
-.text li::before {
- content: "";
- width: 6px;
- height: 1px;
- background: var(--ink);
- flex: none;
- transform: translateY(-4px);
-}
-
-.launch {
- justify-self: start;
- max-width: 100%;
- padding: 8px 12px;
- border: 1px solid var(--line);
- border-radius: var(--radius);
- background: var(--surface);
- font-size: var(--t-12);
- color: var(--ink);
- overflow-wrap: anywhere;
-}
-
-.launch span {
- color: var(--faint);
}
.frame {
- border: 1px solid var(--line);
+ border: 1px solid var(--line-2);
border-radius: var(--radius-lg);
background: var(--surface);
overflow: hidden;
- box-shadow: var(--shadow);
+ box-shadow: var(--shadow), 0 24px 64px -40px color-mix(in srgb, var(--ink) 25%, transparent);
}
.chrome {
display: flex;
align-items: center;
- gap: 6px;
- padding: 10px 14px;
+ gap: var(--s-4);
+ min-height: 44px;
+ padding: 10px 16px;
border-bottom: 1px solid var(--line);
+ background: var(--sunk);
+ font-family: var(--font-mono), ui-monospace, monospace;
+ font-size: var(--t-12);
+ color: var(--muted);
}
-.chrome span:not(:global(.mono)) {
- width: 9px;
- height: 9px;
+.dots {
+ display: flex;
+ gap: 6px;
+}
+
+.dots span {
+ width: 8px;
+ height: 8px;
border-radius: 50%;
- border: 1px solid var(--line-2);
+ background: var(--line-2);
}
-.chrome :global(.mono) {
- margin-left: 10px;
- font-size: var(--t-12);
- color: var(--muted);
+.address {
+ flex: 1;
}
-.shots {
- position: relative;
- background: var(--sunk);
+.recording {
+ white-space: nowrap;
}
.shot {
- position: absolute;
- inset: 0;
width: 100%;
- height: 100%;
- object-fit: cover;
- object-position: top left;
- opacity: 0;
- transform: scale(1.01);
+ height: auto;
filter: none;
- transition: opacity var(--dur-3) var(--ease-out), transform var(--dur-3) var(--ease-out);
}
-.shot[data-on] {
- opacity: 1;
- transform: none;
+.caption {
+ display: flex;
+ align-items: center;
+ justify-content: space-between;
+ gap: var(--s-5);
+ margin-top: var(--s-5);
+ color: var(--muted);
+ font-size: var(--t-12);
+}
+
+.caption ul {
+ display: flex;
+ flex-wrap: wrap;
+ gap: 8px 20px;
+ list-style: none;
+ margin: 0;
+ padding: 0;
}
-@keyframes in {
- from {
- opacity: 0;
- transform: translateY(6px);
- }
- to {
- opacity: 1;
- transform: none;
- }
+.expand {
+ display: inline-flex;
+ align-items: center;
+ gap: 8px;
+ flex-shrink: 0;
+ min-height: 44px;
+ color: var(--ink);
+ text-decoration: underline;
+ text-underline-offset: 4px;
+ text-decoration-color: var(--line-2);
}
-@media (max-width: 900px) {
- .body {
+.expand:hover {
+ color: var(--accent-ink);
+ text-decoration-color: currentColor;
+}
+
+@media (max-width: 720px) {
+ .heading {
grid-template-columns: minmax(0, 1fr);
+ gap: var(--s-4);
+ margin-bottom: var(--s-6);
}
- .frame {
- order: -1;
+
+ .dots {
+ display: none;
}
-}
-@media (prefers-reduced-motion: reduce) {
- .text {
- animation: none;
+ .chrome {
+ padding: 10px 12px;
+ gap: var(--s-2);
}
- .shot {
- transition: none;
+
+ .caption {
+ align-items: start;
+ flex-direction: column;
+ gap: var(--s-2);
+ margin-top: var(--s-4);
}
}
diff --git a/website/components/site/Interfaces.tsx b/website/components/site/Interfaces.tsx
index b173af85..ea561f1c 100644
--- a/website/components/site/Interfaces.tsx
+++ b/website/components/site/Interfaces.tsx
@@ -1,102 +1,47 @@
-"use client";
-
-import { useRef, useState, type KeyboardEvent } from "react";
-import { INTERFACES } from "@/lib/rest";
+import { VIEWER } from "@/lib/rest";
import s from "./Interfaces.module.css";
export function Interfaces() {
- const [active, setActive] = useState(0);
- const [seen, setSeen] = useState<Set<number>>(() => new Set([0]));
- const tabs = useRef<Array<HTMLButtonElement | null>>([]);
- const panel = INTERFACES[active];
-
- const select = (i: number, focus = false) => {
- setActive(i);
- setSeen((prev) => (prev.has(i) ? prev : new Set(prev).add(i)));
- if (focus) tabs.current[i]?.focus();
- };
-
- const onKey = (e: KeyboardEvent<HTMLDivElement>) => {
- const n = INTERFACES.length;
- if (e.key === "ArrowRight") select((active + 1) % n, true);
- else if (e.key === "ArrowLeft") select((active - 1 + n) % n, true);
- else if (e.key === "Home") select(0, true);
- else if (e.key === "End") select(n - 1, true);
- else return;
- e.preventDefault();
- };
-
return (
- <section id="interfaces" className="section" aria-labelledby="interfaces-title">
+ <section id="interfaces" className={`section ${s.section}`} aria-labelledby="interfaces-title">
<div className="wrap">
- <div className="section-head">
- <span className="eyebrow">Interfaces</span>
- <h2 id="interfaces-title">See your agent&apos;s memory.</h2>
- <p>Browse captured observations, memories and sessions from your browser, served from your machine.</p>
- </div>
-
- <div className={s.tabs} role="tablist" aria-label="agentmemory interfaces" onKeyDown={onKey}>
- {INTERFACES.map((it, i) => (
- <button
- key={it.id}
- ref={(el) => {
- tabs.current[i] = el;
- }}
- type="button"
- role="tab"
- id={`tab-${it.id}`}
- aria-selected={i === active}
- aria-controls={`panel-${it.id}`}
- tabIndex={i === active ? 0 : -1}
- className={s.tab}
- onClick={() => select(i)}
- >
- <span>{it.label}</span>
- <span className="mono">{it.port}</span>
- </button>
- ))}
+ <div className={s.heading}>
+ <div>
+ <span className="eyebrow">The AgentMemory viewer</span>
+ <h2 id="interfaces-title">See what your agents remember.</h2>
+ </div>
+ <p>
+ Follow a session, explore connected memories and trace a lesson back to its source.
+ Included with AgentMemory, right on your machine.
+ </p>
</div>
- <div className={s.body}>
- <div className={s.text} role="tabpanel" id={`panel-${panel.id}`} aria-labelledby={`tab-${panel.id}`} key={panel.id}>
- <h3>{panel.title}</h3>
- <p>{panel.body}</p>
- <ul>
- {panel.points.map((p) => (
- <li key={p}>{p}</li>
- ))}
- </ul>
- <code className={s.launch}>
- <span aria-hidden="true">$ </span>
- {panel.launch}
- </code>
- </div>
+ <figure className={s.tour}>
<div className={s.frame}>
- <div className={s.chrome} aria-hidden="true">
- <span />
- <span />
- <span />
- <span className="mono">localhost{panel.port}</span>
- </div>
- <div className={s.shots} style={{ aspectRatio: "16 / 9.1" }}>
- {INTERFACES.map((it, i) =>
- seen.has(i) ? (
- <img
- key={it.id}
- src={it.img}
- alt={it.alt}
- width={it.width}
- height={it.height}
- className={s.shot}
- data-on={i === active || undefined}
- loading={i === 0 ? "lazy" : "eager"}
- decoding="async"
- />
- ) : null,
- )}
+ <div className={s.chrome}>
+ <div className={s.dots} aria-hidden="true"><span /><span /><span /></div>
+ <span className={s.address}>localhost:3113</span>
+ <span className={s.recording}>Recorded tour</span>
</div>
+ <img
+ src={VIEWER.img}
+ alt={VIEWER.alt}
+ width={VIEWER.width}
+ height={VIEWER.height}
+ className={s.shot}
+ loading="lazy"
+ decoding="async"
+ />
</div>
- </div>
+ <figcaption className={s.caption}>
+ <ul aria-label="Viewer features">
+ {VIEWER.features.map((feature) => <li key={feature}>{feature}</li>)}
+ </ul>
+ <a className={s.expand} href={VIEWER.img} target="_blank" rel="noreferrer">
+ Open full-size tour <span aria-hidden="true">↗</span>
+ </a>
+ </figcaption>
+ </figure>
</div>
</section>
);
diff --git a/website/components/site/LocalFirst.tsx b/website/components/site/LocalFirst.tsx
index 18cf0ef8..3e42d42e 100644
--- a/website/components/site/LocalFirst.tsx
+++ b/website/components/site/LocalFirst.tsx
@@ -10,8 +10,8 @@ export function LocalFirst() {
<span className="eyebrow">Local-first</span>
<h2 id="local-title">Everything runs inside your machine.</h2>
<p>
- Capture, storage and search run locally. The REST API and viewer bind to 127.0.0.1 by default. Nothing
- crosses the boundary unless you add a provider yourself.
+ Capture, storage and keyword search run locally. The REST API and viewer bind to 127.0.0.1 by default.
+ Configured remote services receive the text they process. Installation, updates and model downloads need network access.
</p>
</div>
diff --git a/website/lib/bench.ts b/website/lib/bench.ts
index 1438179a..b352d86d 100644
--- a/website/lib/bench.ts
+++ b/website/lib/bench.ts
@@ -1,67 +1,51 @@
-import { LONGMEMEVAL, REPO_URL } from "./site";
-
-export type K = 5 | 10;
-
-export interface TypeHits {
- type: string;
- count: number;
- hits: Record<K, number>;
-}
-
-export const TYPE_HITS: TypeHits[] = LONGMEMEVAL.byType.map((t) => ({
- type: t.type,
- count: t.count,
- hits: { 5: t.h5, 10: t.h10 },
+import evaluation from "./generated-evaluation.json";
+
+export const EVALUATION = evaluation;
+export type Adapter = "agentmemory-bm25-component" | "grep";
+export const ADAPTERS: Adapter[] = ["agentmemory-bm25-component", "grep"];
+export const ADAPTER_LABELS: Record<Adapter, string> = {
+ "agentmemory-bm25-component": "BM25 component",
+ grep: "grep baseline",
+};
+export const TYPE_HITS = Object.entries(evaluation.longmemeval.byType).map(([type, values]) => ({
+ type,
+ count: values.grep.scored,
+ hits: Object.fromEntries(ADAPTERS.map((key) => [key, Math.round(values[key].recallAll * values[key].scored)])) as Record<Adapter, number>,
}));
-
-export function totalHits(k: K): number {
- return TYPE_HITS.reduce((sum, t) => sum + t.hits[k], 0);
+export function totalHits(adapter: Adapter): number {
+ return TYPE_HITS.reduce((sum, row) => sum + row.hits[adapter], 0);
+}
+export function percent(value: number): string {
+ return `${(value * 100).toFixed(1)}%`;
}
-export const benchSrc = {
- readme: `${REPO_URL}/blob/main/benchmark/README.md`,
- budgets: `${REPO_URL}/blob/main/benchmark/capture-costs-budgets.json`,
- costsFigure: `${REPO_URL}/blob/main/docs/benchmarks/capture-costs.svg`,
- longmemevalScript: `${REPO_URL}/blob/main/benchmark/longmemeval-bench.ts`,
- realEmbeddingsScript: `${REPO_URL}/blob/main/benchmark/real-embeddings-eval.ts`,
-};
+export function evaluationProse(data = EVALUATION) {
+ const completion = (rows: Array<{ n: number; completed: number; failed: number }>) => {
+ const totals = rows.reduce((sum, row) => ({
+ n: sum.n + row.n,
+ completed: sum.completed + row.completed,
+ failed: sum.failed + row.failed,
+ }), { n: 0, completed: 0, failed: 0 });
+ return `${totals.completed.toLocaleString("en-US")} of ${totals.n.toLocaleString("en-US")} adapter attempts completed, with ${totals.failed} execution ${totals.failed === 1 ? "failure" : "failures"}.`;
+ };
+ const strictSuccesses = (result: { recallAll: number; scored: number }) =>
+ `${Math.round(result.recallAll * result.scored)}/${result.scored}`;
+ const long = data.longmemeval.byAdapter;
+ const multi = data.longmemeval.byType["multi-session"]["agentmemory-bm25-component"];
+ const coding = data.coding.byAdapter;
+ return {
+ longmemeval: `LongMemEval-S: ${completion(Object.values(long))} ${ADAPTERS.map((key) =>
+ `${ADAPTER_LABELS[key]}: ${long[key].answerable.attempted} answerable and ${long[key].unanswerable} unanswerable questions`
+ ).join("; ")}.`,
+ multiSession: `Multi-session BM25 strict successes: ${strictSuccesses(multi)} (${percent(multi.recallAll)}).`,
+ coding: `Coding-agent-life-v1 strict successes: ${ADAPTERS.map((key) =>
+ `${ADAPTER_LABELS[key]} ${strictSuccesses(coding[key])}`
+ ).join("; ")}. ${completion(Object.values(coding))}`,
+ };
+}
export const REPRODUCE = {
- download: `pip install huggingface_hub
-python3 -c "
-from huggingface_hub import hf_hub_download
-hf_hub_download(repo_id='xiaowu0162/longmemeval-cleaned', filename='longmemeval_s_cleaned.json', repo_type='dataset', local_dir='benchmark/data')
-"`,
- bm25: "npx tsx benchmark/longmemeval-bench.ts bm25",
- hybrid: "npx tsx benchmark/longmemeval-bench.ts hybrid",
- costs: "npm run build && npm run bench:capture-costs",
-};
-
-export const TOKEN_SETUP = {
- observations: 240,
- sessions: 30,
- queries: 20,
- model: "Xenova/all-MiniLM-L6-v2 (384d, local)",
+ download: "mkdir -p benchmark/data\ncurl -fL https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/98d7416c24c778c2fee6e6f3006e7a073259d48f/longmemeval_s_cleaned.json -o benchmark/data/longmemeval_s_cleaned.json\nshasum -a 256 benchmark/data/longmemeval_s_cleaned.json",
+ retrieval: "npm run eval:longmemeval -- --data benchmark/data/longmemeval_s_cleaned.json --adapters grep,bm25",
+ coding: "npm run eval:coding-life -- --adapters grep,bm25",
};
-
-export interface BudgetRow {
- profile: string;
- n: number;
- hookP95Ms: number;
- recoveryIndexReadyMs: number;
- diskPerObsBytes: number;
- idleRssKiB: number;
- searchTokensP50: number;
-}
-
-export const BUDGETS: BudgetRow[] = [
- { profile: "keyless", n: 1000, hookP95Ms: 268, recoveryIndexReadyMs: 3126, diskPerObsBytes: 4510, idleRssKiB: 340760, searchTokensP50: 681 },
- { profile: "keyless", n: 10000, hookP95Ms: 257, recoveryIndexReadyMs: 5742, diskPerObsBytes: 4036, idleRssKiB: 644300, searchTokensP50: 690 },
- { profile: "embed", n: 1000, hookP95Ms: 466, recoveryIndexReadyMs: 5914, diskPerObsBytes: 9552, idleRssKiB: 381080, searchTokensP50: 663 },
- { profile: "embed", n: 10000, hookP95Ms: 406, recoveryIndexReadyMs: 6346, diskPerObsBytes: 9062, idleRssKiB: 863880, searchTokensP50: 678 },
-];
-
-export const RECOVERY_FACTS = [
- { value: "10,000 of 10,000", label: "memories and vectors back after SIGKILL, with 0 embedding calls" },
- { value: "1,000 of 1,000", label: "tool calls kept while the service was down" },
-];
diff --git a/website/lib/compare.ts b/website/lib/compare.ts
index e13c0b12..f9b03549 100644
--- a/website/lib/compare.ts
+++ b/website/lib/compare.ts
@@ -1,6 +1,7 @@
-import { LONGMEMEVAL, REPO_URL } from "@/lib/site";
+import { AGENTS, REPO_URL, src } from "@/lib/site";
+import { EVALUATION, percent } from "@/lib/bench";
-export const CHECKED_ON = "2026-10-05";
+export const CHECKED_ON = "2026-10-28";
export type RowKey =
| "what"
@@ -44,303 +45,373 @@ export interface Competitor {
}
const README = `${REPO_URL}#readme`;
-
export const AGENTMEMORY: Record<RowKey, Fact> = {
- what: {
- value: "Local memory server for coding agents. Hooks capture each session, and memory comes back through hooks and MCP in the next one.",
- src: README,
- },
+ what: { value: "Shared memory for existing coding agents.", src: README },
license: { value: "Apache-2.0", src: `${REPO_URL}/blob/main/LICENSE` },
- deployment: {
- value: "On your machine, in a local iii Worker Compose runtime. The CLI's REST and viewer listeners bind to 127.0.0.1 by default; there is no hosted tier.",
- src: README,
- },
- storage: {
- value: "Nothing to install. State lives in a local data folder. Redis is optional.",
- src: README,
- },
- capture: {
- value: "Automatic. Lifecycle hooks record prompts, tool calls and session ends, so there are no add() calls to write.",
- src: README,
- },
- integrations: {
- value: "25 agents, with native plugins for Claude Code, Codex CLI, Copilot CLI, Cursor and others, 54 MCP tools and a REST API.",
- src: `${REPO_URL}#works-with-every-agent`,
- },
- retrieval: {
- value: "BM25, local vectors and a knowledge graph, fused into one ranking. Works keyless on BM25 alone.",
- src: README,
- },
- benchmarks: {
- value: `LongMemEval-S retrieval recall at 5: ${LONGMEMEVAL.hybrid.r5}%. This measures retrieval only, not answer accuracy.`,
- src: `${REPO_URL}/blob/main/benchmark/LONGMEMEVAL.md`,
- },
- pricing: {
- value: "Free. If you add an LLM or embedding provider, that usage is billed by the provider.",
- src: README,
- },
+ deployment: { value: "Local service; REST and viewer bind to loopback by default.", src: README },
+ storage: { value: "CLI-installed runtime with file-backed state; no separate database service required by default.", src: README },
+ capture: { value: "Automatic capture on supported integrations; explicit MCP and REST writes.", src: README },
+ integrations: { value: `${AGENTS.length} listed agents through native plugins, MCP or REST. Capabilities vary by host.`, src: `${REPO_URL}#works-with-every-agent` },
+ retrieval: { value: "Keyless BM25; optional vectors and graph. Scoped recall, context budgets and source expansion.", src: src.recall },
+ benchmarks: { value: `BM25 component strict recall-all@5: ${percent(EVALUATION.longmemeval.byAdapter["agentmemory-bm25-component"].recallAll)} on 470 answerable LongMemEval-S questions. No reader or daemon.`, src: src.longmemeval },
+ pricing: { value: "No license fee. You operate the runtime; optional hosted providers bill their usage.", src: README },
};
-const MEM0_GH = "https://github.com/mem0ai/mem0";
-const SM_GH = "https://github.com/supermemoryai/supermemory";
-const SM_SELF = "https://supermemory.ai/docs/self-hosting/overview";
-const GRAPHITI_GH = "https://github.com/getzep/graphiti";
-const CM_GH = "https://github.com/thedotmack/claude-mem";
-const LETTA_DOCS = "https://docs.letta.com/";
-const COGNEE_GH = "https://github.com/topoteretes/cognee";
-
export const COMPETITORS: Competitor[] = [
{
- slug: "mem0",
- name: "Mem0",
- tagline: "A memory layer you call from your own agent or app, as a library or a managed platform.",
- links: { home: "https://mem0.ai", docs: "https://docs.mem0.ai", github: MEM0_GH },
- rows: {
- what: { value: "Memory layer for AI agents and apps, offered as an open source library and a managed platform.", src: MEM0_GH },
- license: { value: "Apache-2.0", src: MEM0_GH },
- deployment: { value: "Managed cloud platform, a self-hosted server, or a library inside your app.", src: MEM0_GH },
- storage: { value: "An LLM (OpenAI gpt-5-mini by default) and a vector store. The platform hosts both for you.", src: MEM0_GH },
- capture: { value: "Explicit. Your code calls memory.add() with messages, and an LLM extracts memories from them.", src: MEM0_GH },
- integrations: { value: "Python and TypeScript SDKs, an MCP integration, OpenMemory, and an agent skill for Claude Code.", src: "https://mem0.ai" },
- retrieval: { value: "Multi-signal retrieval over extracted memories. Graph memory is on the Pro and Enterprise plans.", src: "https://mem0.ai/pricing" },
- benchmarks: {
- value: "Their published scores: LoCoMo 92.5, LongMemEval 94.4, BEAM (1M) 64.1. Measured with their own harness and metric, so not comparable to retrieval recall.",
- src: MEM0_GH,
- },
- pricing: { value: "Hobby free (10,000 adds and 1,000 retrievals a month), Starter $19, Pro $249, Enterprise custom.", src: "https://mem0.ai/pricing" },
+ "slug": "mem0",
+ "name": "Mem0",
+ "tagline": "Memory APIs and agent plugins, with hosted and open-source options.",
+ "links": {
+ "home": "https://mem0.ai",
+ "docs": "https://docs.mem0.ai/introduction",
+ "github": "https://github.com/mem0ai/mem0"
},
- summary:
- "Mem0 is a memory API you build into your own application. agentmemory is a memory server for coding agents that captures sessions on its own and runs entirely on your machine.",
- differ: [
- "Mem0 is called by your code: you decide what to send with memory.add(), and an LLM turns it into memories. That suits products where you own the agent loop. agentmemory sits beside agents you already use, like Claude Code or Cursor, and captures their sessions through hooks without code changes.",
- "Mem0 needs an LLM to extract memories and a vector store to hold them, or its managed platform. agentmemory runs as a local service with supervised workers and no external database, and works without any API key on BM25 search.",
- "Some Mem0 capabilities, such as graph memory, on-prem deployment and audit logs, sit on paid plans. agentmemory has no paid tier, so every feature ships in the open source build.",
- ],
- chooseThem: [
- "You are building your own agent or app and want a memory API with Python and TypeScript SDKs.",
- "You want a managed platform with a dashboard and compliance paperwork.",
- "You are happy to choose what gets remembered with explicit add() calls.",
+ "rows": {
+ "what": {
+ "value": "Memory for apps and coding agents",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "license": {
+ "value": "Apache-2.0 library; commercial platform",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "deployment": {
+ "value": "Hosted, self-hosted, or in-process",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "storage": {
+ "value": "Configured models and stores; managed on Platform",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "capture": {
+ "value": "API calls or agent integrations",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "integrations": {
+ "value": "SDKs, MCP, coding-agent plugins",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "retrieval": {
+ "value": "Memory search; graph availability differs between OSS and Platform",
+ "src": "https://docs.mem0.ai/introduction"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://mem0.ai/pricing"
+ }
+ },
+ "summary": "Memory APIs and agent plugins, with hosted and open-source options. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "Mem0 provides application-facing APIs and agent plugins. Check the Platform-versus-OSS feature matrix for the deployment you choose."
],
- chooseUs: [
- "You use existing coding agents and want memory captured without writing code.",
- "You want everything on your machine, with no account and no external database.",
- "You switch between agents and want one memory shared by all of them.",
+ "chooseThem": [
+ "You want a memory API inside an application or a managed platform."
],
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
},
{
- slug: "supermemory",
- name: "supermemory",
- tagline: "A hosted memory and context API with plugins, MCP and framework wrappers.",
- links: { home: "https://supermemory.ai", docs: "https://supermemory.ai/docs", github: SM_GH },
- rows: {
- what: { value: "Memory and context engine for AI, served through a hosted API, plugins and MCP.", src: SM_GH },
- license: {
- value: "MIT for the public repository. Their docs state the self-hosted server binary is built from a separate, non-public codebase.",
- src: SM_SELF,
- },
- deployment: { value: "Hosted API by default. A self-hosted server binary is also available.", src: SM_SELF },
- storage: { value: "Managed by the hosted service. Self-hosting requirements are listed in their docs.", src: SM_SELF },
- capture: { value: "Server-side. Content sent through the API, plugins or MCP is extracted into memories.", src: SM_GH },
- integrations: {
- value: "MCP, a Claude Code plugin, TypeScript and Python SDKs, and wrappers for Vercel AI SDK, LangChain, LangGraph, OpenAI Agents SDK, Mastra and more.",
- src: SM_GH,
- },
- retrieval: { value: "Search over extracted memories and documents through the hosted API.", src: SM_GH },
- benchmarks: {
- value: "They report #1 on LongMemEval (95% recall at 15), LoCoMo and ConvoMem. Recall at 15 is a looser cutoff than recall at 5.",
- src: SM_GH,
- },
- pricing: { value: "Free with $5 of monthly credits, Pro $19, Max $100, Scale $399, Enterprise custom.", src: "https://supermemory.ai/pricing" },
+ "slug": "supermemory",
+ "name": "Supermemory",
+ "tagline": "Memory and document context through hosted or local deployment.",
+ "links": {
+ "home": "https://supermemory.ai",
+ "docs": "https://supermemory.ai/docs/self-hosting/overview",
+ "github": "https://github.com/supermemoryai/supermemory"
},
- summary:
- "supermemory is a hosted memory API with broad framework support. agentmemory is an open source server that runs on your machine and captures coding agent sessions directly.",
- differ: [
- "supermemory runs as a hosted service you send content to, and it extracts memories on its side. agentmemory runs locally, and hooks in your coding agent write to it on your machine.",
- "According to their self-hosting docs, the downloadable server is built from a non-public codebase, collects telemetry unless SUPERMEMORY_DISABLE_TELEMETRY=1 is set, and still calls a hosted reader for URL ingestion. agentmemory is open source end to end and sends nothing anywhere unless you add a provider.",
- "supermemory has wrappers for many app frameworks, which suits product teams. agentmemory focuses on coding agents, with native plugins and hooks for the tools developers already run.",
- ],
- chooseThem: [
- "You want a managed API and do not want to run anything yourself.",
- "You build on Vercel AI SDK, LangChain, Mastra or similar and want a drop-in wrapper.",
- "You need memory and document retrieval for an end-user product.",
+ "rows": {
+ "what": {
+ "value": "Memory and document context",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "license": {
+ "value": "Local server code is non-public; SDKs are public",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "deployment": {
+ "value": "Hosted API or local server binary",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "storage": {
+ "value": "Embedded storage and local embeddings in the local server",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "capture": {
+ "value": "API, MCP or coding-agent plugins",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "integrations": {
+ "value": "SDKs and agent integrations",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "retrieval": {
+ "value": "Memory and document retrieval",
+ "src": "https://supermemory.ai/docs/self-hosting/overview"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://supermemory.ai/pricing"
+ }
+ },
+ "summary": "Memory and document context through hosted or local deployment. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "Supermemory supports local deployment. Its local server and hosted API have different feature sets and licensing; the local server is a separate non-public codebase."
],
- chooseUs: [
- "You want the whole engine open source and auditable.",
- "You want zero network calls by default, with no telemetry to switch off.",
- "Your main use is coding agents like Claude Code, Codex or Cursor.",
+ "chooseThem": [
+ "You want application context APIs or its document integrations."
],
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
},
{
- slug: "zep",
- name: "Zep",
- tagline: "An enterprise context layer built on Graphiti, an open source temporal knowledge graph.",
- links: { home: "https://www.getzep.com", docs: "https://help.getzep.com", github: GRAPHITI_GH },
- rows: {
- what: {
- value: "Context layer for enterprise agents that unifies business data, documents and conversations. Built on Graphiti, its open source temporal knowledge graph.",
- src: "https://www.getzep.com",
- },
- license: { value: "Zep is a commercial service. Graphiti is Apache-2.0.", src: GRAPHITI_GH },
- deployment: { value: "Zep Cloud, with bring-your-own-keys and bring-your-own-cloud options. Graphiti runs wherever you host it.", src: "https://www.getzep.com" },
- storage: { value: "Graphiti needs a graph database (Neo4j, FalkorDB or Amazon Neptune) and an LLM that supports structured output.", src: GRAPHITI_GH },
- capture: { value: "Explicit. Your application sends conversations and data to Zep, or adds them to Graphiti through its API.", src: GRAPHITI_GH },
- integrations: { value: "A memory MCP server for Zep. Graphiti ships its own MCP server.", src: "https://www.getzep.com" },
- retrieval: { value: "Hybrid semantic, BM25 keyword and graph traversal over a temporal knowledge graph.", src: GRAPHITI_GH },
- benchmarks: {
- value: "Their published scores: LoCoMo 94.7% and LongMemEval 90.2% accuracy, with 155 to 162 ms retrieval latency. Answer accuracy, not retrieval recall.",
- src: "https://www.getzep.com",
- },
- pricing: { value: "10,000 free credits a month, Flex $125 a month, Flex Plus $375 a month, Enterprise custom.", src: "https://www.getzep.com/pricing/" },
+ "slug": "zep",
+ "name": "Zep",
+ "tagline": "Enterprise context, alongside the open-source Graphiti framework.",
+ "links": {
+ "home": "https://www.getzep.com",
+ "docs": "https://github.com/getzep/graphiti",
+ "github": "https://github.com/getzep/graphiti"
},
- summary:
- "Zep is a managed context platform for enterprise agents, built on the Graphiti graph engine. agentmemory is a local memory server for individual developers and their coding agents.",
- differ: [
- "Zep centers on a temporal knowledge graph that combines business data, documents and conversations for many users. agentmemory centers on one developer's coding sessions, captured automatically on their machine.",
- "Running Graphiti yourself means operating a graph database and an LLM for ingestion. agentmemory needs neither: its local iii Worker Compose runtime manages the services, and its own graph is built locally and optionally.",
- "Zep's platform adds enterprise controls such as SOC 2 Type II and bring-your-own-cloud. agentmemory avoids that surface by not hosting anything, since data never leaves the machine.",
- ],
- chooseThem: [
- "You need governed, shared context across many users and business systems.",
- "You want a managed graph platform with enterprise compliance.",
- "You are already invested in Neo4j or Neptune and want Graphiti on top.",
+ "rows": {
+ "what": {
+ "value": "Enterprise context; Graphiti temporal graphs",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "license": {
+ "value": "Commercial Zep; Apache-2.0 Graphiti",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "deployment": {
+ "value": "Managed or BYOC Zep; self-hosted Graphiti",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "storage": {
+ "value": "Zep manages its stack; Graphiti needs configured database/models",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "capture": {
+ "value": "Application ingestion",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "integrations": {
+ "value": "SDKs and MCP",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "retrieval": {
+ "value": "Temporal graph and hybrid retrieval",
+ "src": "https://github.com/getzep/graphiti"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://www.getzep.com/pricing/"
+ }
+ },
+ "summary": "Enterprise context, alongside the open-source Graphiti framework. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "Zep and Graphiti are distinct deployment choices. Evaluate Zep for enterprise context and Graphiti for a graph framework you operate."
],
- chooseUs: [
- "You want memory for your own coding agents without running a graph database.",
- "You want capture to happen through hooks, with no integration code.",
- "You want it free and local, with nothing to provision.",
+ "chooseThem": [
+ "You need shared business context or temporal graph infrastructure."
],
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
},
{
- slug: "claude-mem",
- name: "claude-mem",
- tagline: "A memory plugin that compresses coding agent sessions with Claude and injects them back.",
- links: { home: CM_GH, docs: "https://docs.claude-mem.ai/", github: CM_GH },
- rows: {
- what: {
- value: "Memory plugin that captures coding agent sessions, compresses them with AI and injects relevant context into future sessions.",
- src: CM_GH,
- },
- license: { value: "Apache-2.0", src: CM_GH },
- deployment: { value: "Runs locally next to your agent, with a background worker and a web viewer.", src: CM_GH },
- storage: { value: "SQLite with FTS5 and a Chroma vector database. Compression goes through the Claude Agent SDK, so it needs Claude access.", src: CM_GH },
- capture: { value: "Automatic, through 5 lifecycle hooks: SessionStart, UserPromptSubmit, PostToolUse, Stop and SessionEnd.", src: CM_GH },
- integrations: {
- value: "Built for Claude Code. The README also lists OpenClaw, Codex, Gemini, Hermes, Copilot and OpenCode, and it exposes MCP search tools.",
- src: CM_GH,
- },
- retrieval: { value: "Hybrid semantic and keyword search (Chroma plus SQLite FTS5) through a search, timeline and get_observations flow.", src: CM_GH },
- benchmarks: { value: "not documented", src: CM_GH },
- pricing: { value: "Free and open source. Compression uses your Claude usage.", src: CM_GH },
+ "slug": "claude-mem",
+ "name": "claude-mem",
+ "tagline": "Coding-session capture with model-backed summaries.",
+ "links": {
+ "home": "https://github.com/thedotmack/claude-mem",
+ "docs": "https://github.com/thedotmack/claude-mem",
+ "github": "https://github.com/thedotmack/claude-mem"
},
- summary:
- "claude-mem and agentmemory solve the same problem the same way: hooks capture coding sessions and context comes back next time. They differ in what runs the compression, what storage they need, and how much of memory they model.",
- differ: [
- "claude-mem compresses observations with Claude through the Claude Agent SDK, so every compression is a model call. agentmemory stores observations as captured and works with no model at all; an LLM is optional and only used for consolidation if you add one.",
- "claude-mem uses SQLite plus a Chroma vector database. agentmemory runs on a local iii Worker Compose runtime with file-backed state by default and local embeddings, and adds a knowledge graph and four memory tiers with decay.",
- "Both capture through hooks. agentmemory also ships native plugins for Codex CLI, Cursor, Copilot CLI and others, 54 MCP tools, a REST API, and a release gate that checks crash recovery on every publish.",
- ],
- chooseThem: [
- "You live in Claude Code and want AI-written summaries of every session.",
- "You are fine spending Claude usage on compression.",
- "You prefer its search, timeline and observation flow.",
- ],
- chooseUs: [
- "You want memory that works with no model calls and no API key.",
- "You use several agents and want one shared memory across them.",
- "You want tiered memory with decay, a knowledge graph and crash-safe capture.",
+ "rows": {
+ "what": {
+ "value": "Coding-session memory",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "license": {
+ "value": "Apache-2.0 project",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "deployment": {
+ "value": "Local worker; hosted observer option",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "storage": {
+ "value": "Worker storage and a selected summary provider",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "capture": {
+ "value": "Native integrations and hooks",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "integrations": {
+ "value": "Multiple coding agents, including Pi and OMP",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "retrieval": {
+ "value": "Search, timeline and source observations",
+ "src": "https://github.com/thedotmack/claude-mem"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://github.com/thedotmack/claude-mem"
+ }
+ },
+ "summary": "Coding-session capture with model-backed summaries. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "claude-mem offers several summary providers, including a hosted observer. AgentMemory also offers optional LLM processing, while its keyless mode stores and searches captures without a model."
],
- moving: [
- "Install agentmemory with npx -y @agentmemory/agentmemory@latest and connect Claude Code during setup.",
- "Bring in past sessions from your Claude Code transcripts with npx -y @agentmemory/agentmemory@latest import-jsonl. Claude Code keeps transcripts for 30 days by default.",
- "Disable the claude-mem plugin so only one tool captures each session.",
+ "chooseThem": [
+ "You prefer its observer and session-summary workflow."
],
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
},
{
- slug: "letta",
- name: "Letta",
- tagline: "A platform for stateful agents, where the agent edits its own memory.",
- links: { home: "https://www.letta.com", docs: LETTA_DOCS, github: "https://github.com/letta-ai/letta" },
- rows: {
- what: {
- value: "Platform for building stateful agents. An agent runtime with memory, rather than memory for an agent you already use.",
- src: LETTA_DOCS,
- },
- license: { value: "Apache-2.0", src: "https://github.com/letta-ai/letta" },
- deployment: { value: "Letta Cloud, or the Letta server run yourself.", src: LETTA_DOCS },
- storage: { value: "Handled by Letta Cloud or the self-hosted Letta server.", src: LETTA_DOCS },
- capture: {
- value: "Agent-managed. The agent reads and updates its own memory blocks with built-in memory tools.",
- src: "https://docs.letta.com/v1-sdk/memory/memory-blocks",
- },
- integrations: {
- value: "Python and TypeScript SDKs. Letta Code is its open source coding agent harness.",
- src: "https://docs.letta.com/v1-sdk/memory/memory-blocks",
- },
- retrieval: {
- value: "Core memory blocks are injected into the context window, and the agent changes them through tools.",
- src: "https://docs.letta.com/v1-sdk/concepts/stateful-agents",
- },
- benchmarks: { value: "not documented", src: LETTA_DOCS },
- pricing: { value: "not documented", src: LETTA_DOCS },
+ "slug": "letta",
+ "name": "Letta",
+ "tagline": "A stateful-agent runtime with persistent memory.",
+ "links": {
+ "home": "https://www.letta.com",
+ "docs": "https://docs.letta.com/",
+ "github": "https://github.com/letta-ai/letta-code"
},
- summary:
- "Letta is a runtime you build agents in, and its agents manage their own memory. agentmemory adds memory to agents you already use, without changing how they run.",
- differ: [
- "With Letta you create agents on its platform, and each agent edits its memory blocks through tools. With agentmemory you keep your current coding agent, and memory is captured around it by hooks.",
- "Letta memory belongs to Letta agents. agentmemory memory is shared by every connected agent, so Claude Code, Cursor and Codex all read from one store.",
- "Letta runs as Letta Cloud or a Letta server. agentmemory runs locally through Worker Compose, which manages its engine and workers.",
- ],
- chooseThem: [
- "You are building your own long-running agents and want a full runtime.",
- "You want agents that decide for themselves what to remember.",
- "You want shared memory blocks across agents you design.",
+ "rows": {
+ "what": {
+ "value": "Stateful-agent platform and harness",
+ "src": "https://docs.letta.com/"
+ },
+ "license": {
+ "value": "Open-source harness; commercial cloud options",
+ "src": "https://docs.letta.com/"
+ },
+ "deployment": {
+ "value": "Managed cloud or self-hosted runtime",
+ "src": "https://docs.letta.com/"
+ },
+ "storage": {
+ "value": "Agent runtime and configured models",
+ "src": "https://docs.letta.com/"
+ },
+ "capture": {
+ "value": "Agent-managed memory",
+ "src": "https://docs.letta.com/"
+ },
+ "integrations": {
+ "value": "CLI, Agent SDK, ACP and MCP",
+ "src": "https://docs.letta.com/"
+ },
+ "retrieval": {
+ "value": "MemFS and shared memory",
+ "src": "https://docs.letta.com/"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://docs.letta.com/"
+ }
+ },
+ "summary": "A stateful-agent runtime with persistent memory. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "Letta supplies the agent runtime and its memory system. AgentMemory supplies a shared memory service for tools you already use."
],
- chooseUs: [
- "You want memory for the coding agents you already use.",
- "You want capture to be automatic instead of agent-decided.",
- "You want a local service with automatic worker supervision.",
+ "chooseThem": [
+ "You want to build or run stateful agents in an integrated harness."
],
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
},
{
- slug: "cognee",
- name: "Cognee",
- tagline: "An open source memory platform that turns your data into a knowledge graph.",
- links: { home: "https://www.cognee.ai", docs: "https://docs.cognee.ai/", github: COGNEE_GH },
- rows: {
- what: { value: "Open source AI memory platform for agents that builds a knowledge graph from the data you give it.", src: COGNEE_GH },
- license: { value: "Apache-2.0", src: COGNEE_GH },
- deployment: { value: "Python library, Docker image, or Cognee Cloud.", src: COGNEE_GH },
- storage: { value: "Graph, vector and relational stores. It can run the whole memory layer on a single Postgres instance.", src: COGNEE_GH },
- capture: {
- value: "Explicit. Your code calls cognee.remember (or add), and a pipeline turns the data into a graph. An LLM is optional with a local extraction model.",
- src: COGNEE_GH,
- },
- integrations: { value: "MCP server, a Claude Code memory plugin, Cursor and Cline, plus Python, TypeScript and Rust clients and REST.", src: COGNEE_GH },
- retrieval: { value: "cognee.recall routes between graph, vector and code retrieval, or combines them.", src: COGNEE_GH },
- benchmarks: {
- value: "Their published scores: 0.79 at 100K tokens and 0.67 at 10M tokens of context on their own evaluation.",
- src: COGNEE_GH,
- },
- pricing: { value: "The library is free. Cognee Cloud pricing: not documented.", src: COGNEE_GH },
+ "slug": "cognee",
+ "name": "Cognee",
+ "tagline": "Knowledge graphs from documents, code and conversations.",
+ "links": {
+ "home": "https://www.cognee.ai",
+ "docs": "https://github.com/topoteretes/cognee",
+ "github": "https://github.com/topoteretes/cognee"
},
- summary:
- "Cognee builds knowledge graphs from data you feed it, for agents and apps. agentmemory captures coding agent sessions automatically and keeps them on your machine.",
- differ: [
- "Cognee is a pipeline you run over documents and data: add them, then turn them into a graph. agentmemory captures what your coding agent does as it works, through hooks.",
- "Cognee relies on graph, vector and relational stores, which can share one Postgres. agentmemory uses its iii state worker with file-backed key/value storage by default, managed by a local Worker Compose runtime.",
- "Both offer MCP and a Claude Code integration. agentmemory adds native plugins for many coding agents, a live viewer, and four memory tiers with decay.",
- ],
- chooseThem: [
- "You want to build a knowledge graph from documents, code and data sources.",
- "You need Python, TypeScript and Rust clients in your own app.",
- "You already run Postgres and want memory inside it.",
+ "rows": {
+ "what": {
+ "value": "Memory platform and knowledge pipeline",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "license": {
+ "value": "Open-source project; commercial offerings",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "deployment": {
+ "value": "Local library/server or Cognee Cloud",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "storage": {
+ "value": "Configured storage; local extraction/embedding models available",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "capture": {
+ "value": "API, CLI, plugins and MCP",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "integrations": {
+ "value": "Python, TypeScript, Rust and coding-agent integrations",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "retrieval": {
+ "value": "Graph, vector and code retrieval",
+ "src": "https://github.com/topoteretes/cognee"
+ },
+ "benchmarks": {
+ "value": "No matched head-to-head run in this evaluation.",
+ "src": "https://github.com/rohitg00/agentmemory/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md"
+ },
+ "pricing": {
+ "value": "See current plans and provider costs.",
+ "src": "https://docs.cognee.ai/cognee-cloud/overview"
+ }
+ },
+ "summary": "Knowledge graphs from documents, code and conversations. AgentMemory provides local memory shared by existing coding agents.",
+ "differ": [
+ "Cognee builds connected knowledge from varied data and supports a keyless local path. AgentMemory focuses on captured coding history, scoped recall and a shared viewer."
],
- chooseUs: [
- "You want your coding sessions remembered without feeding data in by hand.",
- "You want no database to run.",
- "You want one memory shared across many coding agents.",
+ "chooseThem": [
+ "You want a knowledge pipeline across documents, code and conversations."
],
- },
+ "chooseUs": [
+ "You want shared coding history with scoped recall, source expansion and a local viewer.",
+ "You want keyword retrieval without an API key."
+ ]
+ }
];
export function getCompetitor(slug: string) {
diff --git a/website/lib/generated-changelog.json b/website/lib/generated-changelog.json
index 2c3a61d8..8a85ae1e 100644
--- a/website/lib/generated-changelog.json
+++ b/website/lib/generated-changelog.json
@@ -1,4 +1,9 @@
[
+ {
+ "version": "0.9.31",
+ "date": "2026-10-25",
+ "body": "### Upgrade notes\n\n- **The runtime moves to iii 0.25.0 and Worker Compose.** Stop the existing instance with its old CLI before upgrading, then restart with the same instance, port and data directory. File state, the daemon secret and Docker data volumes are preserved. The CLI no longer uses `AGENTMEMORY_USE_DOCKER`, `AGENTMEMORY_III_CONFIG` or custom `iii-config.yaml` files; Docker deployments use the shared deployment image. Follow the [upgrade guide](docs/upgrades/iii-0.25.md) for these installation changes.\n- **Viewer updates use authenticated SSE** on the REST port at `/agentmemory/viewer/events`; port 3112 is unused. `agentmemory console` now prints migration guidance and the viewer URL. The viewer remains on port 3113 by default. Status reports the configured engine pin as `service.enginePinnedVersion` instead of `engineVersion`.\n- **Narrative recall now returns structured JSON through MCP.** Read the `text` field for the narrative and the accompanying budget metadata for omitted or shortened results. See [recall guidance](docs/recall.md).\n- Full historical vector recovery requires `AGENTMEMORY_VECTOR_BACKFILL=all` on each start and processes at most `AGENTMEMORY_VECTOR_BACKFILL_MAX` missing records per start. Keyword search remains available while recovery is paused. See [vector recovery](docs/reliability/source-and-vector-recovery.md) before rebuilding an old index.\n- Re-run `agentmemory connect pi --with-mcp` or `agentmemory connect omp --with-mcp` after upgrading to refresh their bundled paths and skills. Existing DSH hook-bridge installations migrate with `agentmemory connect dsh --with-hooks --force`; the connector backs up configuration and preserves unrelated entries.\n\n### Added\n\n- **Installable Pi package and native OMP integration.** Pi's `@agentmemory/pi` package contains the compiled extension and shared skills. Both connectors install project recall, capture, native memory tools and optional MCP, with bounded offline capture queues. The connector works before the separate Pi package's first npm publication.\n- **Native DSH capture** ships inside AgentMemory, with prompt, tool and assistant-reply capture, privacy filtering and offline recovery. No separate capture package is needed (#1609, including instruction-file setup by hejiawow).\n- **Cursor workspace resolution** maps lifecycle hooks to the correct project using validated paths and local session metadata, with detached stop/session-end delivery (#1112).\n- **Paged exports** support `collectionLimit`, `collectionOffset` and `collections` through REST and MCP. Partial exports cannot be imported with `strategy: \"replace\"`.\n- **Compression recovery** preserves searchable captures after optional model compression fails. `agentmemory compression` inspects failures and supports scoped retries; summary repair resumes after late capture and restart.\n\n### Fixed\n\n- Saved-memory recall preserves project and agent scope, supports memory-only retrieval and retains recovery references when a token budget omits content. Graph retrieval uses bounded shared reads and keeps each observation's best score; optional reranking uses the model's actual paired-query scores.\n- Vector recovery retains unfinished work across checkpoints, provider failures and restarts. Redis state mutations serialize per key within a worker, and deletion keeps graph and facet indexes consistent.\n- Imports validate collection shapes and row keys before a replace operation deletes stored data. The import observation ceiling no longer changes with the ordinary capture limit.\n- Hooks capture final assistant replies and distinguish separate subagent, task and notification events. Structured redaction preserves identifiers and retained custom-event content remains searchable.\n- MCP keeps routine logs off stdout and reports when a write only reached its local fallback store. The viewer loads actions beyond the first page.\n- **JSONL import keeps real session times and no longer duplicates on re-import.** `parseJsonlText` took `startedAt`/`endedAt` from the first and last line even when it was a timestampless metadata entry (`last-prompt`, `cost-state`), so the session got the import time. It now uses only entries that carry a timestamp. Observation ids are derived from the entry `uuid` (or session id and line position when absent) instead of a random id, so importing the same transcript again overwrites the same rows, and `observationCount` counts only rows that were new. Rows imported by earlier versions keep their random ids, so the first re-import after upgrading adds one last copy.\n- **Session summaries cover the whole session again** (#745). 0.9.30 completed a session on the first per-turn Stop and ignored every later end, so summaries and graph extraction stopped at turn 1. `mem::observe` now reopens a completed session when new observations arrive, the Stop hook sends `final: false` so the session stays open, and an idle finalizer completes it once nothing has happened for `AGENTMEMORY_FINALIZE_IDLE_MS` (default 2 minutes). SessionEnd, the CLI and the OpenCode plugin send `final: true` and complete at once; a request without the flag behaves as before. The sweep summarizes a session before marking it abandoned (#737).\n- **The Devin plugin captures sessions.** A plugin install loaded `hooks/hooks.json`, whose capitalized Claude Code matchers never match Devin's tool names and whose extra events Devin rejects, so capture never fired. `plugin/hooks.json` now carries the six Devin hooks with lowercase matchers, and Devin loads it first. Verified on Devin CLI 3000.11.3: one session recorded session start, prompt, tool use and session end.\n- **Summaries are incremental.** `mem::summarize` keeps a watermark (`coveredThrough`) and summarizes only the observations captured since the last run, merging them into the stored summary with one reduce call. A session end with nothing new costs no LLM call, and the reflect, graph and skill fan-out is skipped for it.\n\n### Changed\n\n- **Publishing verifies the exact gated tarballs.** The workflow checks SHA512 integrity and signed provenance against the source repository, workflow, ref and commit before accepting an existing version or completing publication. Conflicting artifacts fail instead of being skipped.\n- Shared agent skills now describe the current runtime, credential discovery, native integrations and capture limits. The README and translations link to current setup, storage and upgrade guidance.\n- The Devin plugin manifest adds a display name and logo, says it is for Devin CLI and Devin Desktop only, and runs `@agentmemory/mcp` at the release version instead of the latest one. `npm run docs:sync` moves that pin on each version bump.\n- `@agentmemory/mcp` pins the exact matching `@agentmemory/agentmemory` release instead of a `~` range, so an installed shim never resolves a newer runtime than the one it was published with."
+ },
{
"version": "0.9.30",
"date": "2026-10-06",
@@ -53,10 +58,5 @@
"version": "0.9.20",
"date": "2026-05-18",
"body": "Hotfix: revert the Codex Stop → session-end chain shipped in v0.9.19.\n\n### Fixed\n\n- **Revert Codex Stop hook session-end chain** ([PR #501](https://github.com/rohitg00/agentmemory/pull/501) by [@Rex57](https://github.com/Rex57), reverts v0.9.19's [#495](https://github.com/rohitg00/agentmemory/pull/495), re-opens [#493](https://github.com/rohitg00/agentmemory/issues/493)). Post-merge field-testing surfaced the underlying issue: Codex `Stop` fires before the overall conversation is truly finished — multiple Stops bracket each assistant turn within one session. Chaining `session-end.mjs` from Stop marked sessions completed too early, with later observations still arriving against an `endedAt`-stamped record. Restored to summarize-only Stop; the SessionEnd-shaped solution stays open as [#493](https://github.com/rohitg00/agentmemory/issues/493)."
- },
- {
- "version": "0.9.19",
- "date": "2026-05-18",
- "body": "Feature + hardening wave. Sessions now link to the git commits they shipped (forward + reverse lookup, REST + MCP surfaces). OpenAI provider transport collapses into one shared module with auto-detected Azure URL style (legacy `/openai/deployments/<dep>` and v1 `/openai/v1` both supported). Graph retrieval switches from BFS to Dijkstra over the weighted edge graph. Codex Stop hook chains session-end. Plugin MCP server inherits `AGENTMEMORY_URL` / `AGENTMEMORY_SECRET` from the shell. Point fix routes the bundled iii-console installer around an upstream tag-prefix bug. 1007+ tests pass.\n\n### Added\n\n- **Session-to-commit linking** ([PR #498](https://github.com/rohitg00/agentmemory/pull/498)). New `KV.commits` namespace keyed by full SHA holds `CommitLink` records (sha, shortSha, branch, repo, message, author, authoredAt, files, sessionIds). `Session.commitShas[]` provides the forward back-reference. REST: `POST /agentmemory/session/commit` upserts links (merges `sessionIds` on re-link, preserves `linkedAt`); `GET /agentmemory/commits/:sha` and `GET /agentmemory/commits` round-trip. MCP: `memory_commit_lookup` and `memory_commits` tools. Post-commit hook auto-captures the link on every commit in the working directory. Closes the loop on \"what session wrote this code\" / \"what commits did this session ship\" without leaving agentmemory.\n\n- **Azure OpenAI v1 URL pattern auto-detection** ([PR #462](https://github.com/rohitg00/agentmemory/pull/462), closes [#371](https://github.com/rohitg00/agentmemory/issues/371)). Both the LLM and embedding providers now route through `_openai-shared.ts` and auto-detect Azure URL style:\n - `OPENAI_BASE_URL=https://r.openai.azure.com/openai/deployments/<dep>` → legacy URL pattern, `api-version` query param via `OPENAI_API_VERSION` (default `2024-08-01-preview`).\n - `OPENAI_BASE_URL=https://r.openai.azure.com` (bare host, or `/openai`, or `/openai/v1`) → v1 GA pattern, `/openai/v1/<route>`, deployment name carried in the request body as `model`.\n - Net effect: Azure embeddings work for the first time (LLM-side Azure shipped in v0.9.17; embedding was still hardcoded to `/v1/embeddings` + Bearer). Closes [#199](https://github.com/rohitg00/agentmemory/issues/199) (consolidation) as superseded.\n\n### Changed\n\n- **Graph retrieval: BFS → Dijkstra over weighted edges** ([PR #463](https://github.com/rohitg00/agentmemory/pull/463), closes [#328](https://github.com/rohitg00/agentmemory/issues/328) filed by [@Tanmay-008](https://github.com/Tanmay-008) with benchmark numbers showing the BFS edge-count semantics + O(n) `Array.shift()` profile). Memory graph edges carry weights 0.1–1.0; BFS visited by edge-count regardless, so a one-hop weak edge ranked the same as a two-hop strong chain. Dijkstra over `cost = 1/max(weight, 0.01)` selects weight-optimal paths instead. Adjacency map built once in O(V+E) and min-heap dequeue at O(log V) replace the prior O(V·E) + O(n) `Array.shift()` profile. `maxDepth` semantics preserved (still edge-count bound). The `startNode` path is excluded from returned paths so the dedicated `score=1.0` fallback loop in `searchByEntities` fires as designed (regression catch from the inline review).\n\n- **Codex Stop hook chains session-end** ([PR #495](https://github.com/rohitg00/agentmemory/pull/495) by [@Rex57](https://github.com/Rex57), fixes [#493](https://github.com/rohitg00/agentmemory/issues/493) also filed by [@Rex57](https://github.com/Rex57)). `hooks.codex.json` Stop now runs `stop.mjs` followed by `session-end.mjs`. Codex doesn't ship a separate `SessionEnd` lifecycle event, so the session would otherwise stay open after the user terminated the Codex session. Tests assert both commands appear in the Stop chain.\n\n- **Plugin MCP entry inherits shell env via passthrough** ([PR #460](https://github.com/rohitg00/agentmemory/pull/460), closes [#375](https://github.com/rohitg00/agentmemory/issues/375) filed by [@anthony-spruyt](https://github.com/anthony-spruyt)). `plugin/.mcp.json` and `AGENTMEMORY_MCP_BLOCK` (the template `agentmemory connect <agent>` writes into `~/.claude.json`, `~/.cursor/mcp.json`, etc.) now declare `env: { AGENTMEMORY_URL: \"${AGENTMEMORY_URL}\", AGENTMEMORY_SECRET: \"${AGENTMEMORY_SECRET}\" }`. The MCP host substitutes shell values at server launch. When the vars are unset, the host passes empty strings; the standalone shim's `process.env[\"AGENTMEMORY_URL\"] || \"http://localhost:3111\"` falls back to localhost. One wired entry now covers both local and remote (k8s / reverse-proxied) deployments without `/doctor` \"duplicate server\" warnings.\n\n### Fixed\n\n- **`ensureIiiConsole()` install path** (`src/cli.ts`). The upstream `install.iii.dev/console/main/install.sh` script's jq predicate filters releases with `startswith(\"v\")` while `iii-hq/iii` tags as `iii/v0.12.0`. The script bails with `no stable iii release found` on every fresh install. Switched the install command to `curl ... | bash -s -- --next` until upstream patches the script — the `--next` codepath uses a regex on `-next.` without the buggy `startswith` constraint, so it succeeds against the same tag set. Inline comment documents the upstream bug + revert condition.\n\n### Infrastructure\n\n- 91 test files, 1007 tests pass on `chore(release): v0.9.19`.\n- Bundles five PRs ([#460](https://github.com/rohitg00/agentmemory/pull/460), [#462](https://github.com/rohitg00/agentmemory/pull/462), [#463](https://github.com/rohitg00/agentmemory/pull/463), [#495](https://github.com/rohitg00/agentmemory/pull/495), [#498](https://github.com/rohitg00/agentmemory/pull/498)) plus the inline iii-console installer workaround into one patch release."
}
]
diff --git a/website/lib/generated-evaluation.json b/website/lib/generated-evaluation.json
new file mode 100644
index 00000000..e4e53f6b
--- /dev/null
+++ b/website/lib/generated-evaluation.json
@@ -0,0 +1,986 @@
+{
+ "longmemeval": {
+ "byAdapter": {
+ "grep": {
+ "n": 500,
+ "completed": 500,
+ "failed": 0,
+ "scored": 470,
+ "answerable": {
+ "attempted": 470,
+ "completed": 470,
+ "failed": 0
+ },
+ "unanswerable": 30,
+ "unanswerableCompleted": 30,
+ "unanswerableFailed": 0,
+ "p": 0.2953191489361707,
+ "r": 0.8081914893617022,
+ "recallAll": 0.7212765957446808,
+ "hit": 417,
+ "strictSuccessOverAttempts": 0.7212765957446808,
+ "latencyP50": 2.352957999999944,
+ "latencyP95": 5.313833000000159,
+ "ingestionP50": 0.002499999999827196
+ },
+ "agentmemory-bm25-component": {
+ "n": 500,
+ "completed": 500,
+ "failed": 0,
+ "scored": 470,
+ "answerable": {
+ "attempted": 470,
+ "completed": 470,
+ "failed": 0
+ },
+ "unanswerable": 30,
+ "unanswerableCompleted": 30,
+ "unanswerableFailed": 0,
+ "p": 0.3370212765957453,
+ "r": 0.9171985815602839,
+ "recallAll": 0.8468085106382979,
+ "hit": 452,
+ "strictSuccessOverAttempts": 0.8468085106382979,
+ "latencyP50": 1.752375000000029,
+ "latencyP95": 2.0167920000021695,
+ "ingestionP50": 128.29283299999952
+ }
+ },
+ "byType": {
+ "single-session-user": {
+ "grep": {
+ "n": 70,
+ "completed": 70,
+ "failed": 0,
+ "scored": 64,
+ "answerable": {
+ "attempted": 64,
+ "completed": 64,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.18749999999999983,
+ "r": 0.9375,
+ "recallAll": 0.9375,
+ "hit": 60,
+ "strictSuccessOverAttempts": 0.9375,
+ "latencyP50": 1.6280829999998332,
+ "latencyP95": 2.776290999999901,
+ "ingestionP50": 0.0018749999999272404
+ },
+ "agentmemory-bm25-component": {
+ "n": 70,
+ "completed": 70,
+ "failed": 0,
+ "scored": 64,
+ "answerable": {
+ "attempted": 64,
+ "completed": 64,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.1968749999999998,
+ "r": 0.984375,
+ "recallAll": 0.984375,
+ "hit": 63,
+ "strictSuccessOverAttempts": 0.984375,
+ "latencyP50": 1.7266249999993306,
+ "latencyP95": 1.9502500000007785,
+ "ingestionP50": 127.6830000000009
+ }
+ },
+ "multi-session": {
+ "grep": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 121,
+ "answerable": {
+ "attempted": 121,
+ "completed": 121,
+ "failed": 0
+ },
+ "unanswerable": 12,
+ "unanswerableCompleted": 12,
+ "unanswerableFailed": 0,
+ "p": 0.37355371900826406,
+ "r": 0.7450413223140495,
+ "recallAll": 0.5289256198347108,
+ "hit": 114,
+ "strictSuccessOverAttempts": 0.5289256198347108,
+ "latencyP50": 2.219832999999653,
+ "latencyP95": 3.6175829999997404,
+ "ingestionP50": 0.0025840000002972374
+ },
+ "agentmemory-bm25-component": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 121,
+ "answerable": {
+ "attempted": 121,
+ "completed": 121,
+ "failed": 0
+ },
+ "unanswerable": 12,
+ "unanswerableCompleted": 12,
+ "unanswerableFailed": 0,
+ "p": 0.44132231404958633,
+ "r": 0.8701101928374656,
+ "recallAll": 0.7107438016528925,
+ "hit": 117,
+ "strictSuccessOverAttempts": 0.7107438016528925,
+ "latencyP50": 1.7304590000021562,
+ "latencyP95": 1.9607500000001892,
+ "ingestionP50": 128.9686669999992
+ }
+ },
+ "single-session-preference": {
+ "grep": {
+ "n": 30,
+ "completed": 30,
+ "failed": 0,
+ "scored": 30,
+ "answerable": {
+ "attempted": 30,
+ "completed": 30,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.09333333333333334,
+ "r": 0.4666666666666667,
+ "recallAll": 0.4666666666666667,
+ "hit": 14,
+ "strictSuccessOverAttempts": 0.4666666666666667,
+ "latencyP50": 2.378375000000233,
+ "latencyP95": 3.53387500000008,
+ "ingestionP50": 0.002540999999837368
+ },
+ "agentmemory-bm25-component": {
+ "n": 30,
+ "completed": 30,
+ "failed": 0,
+ "scored": 30,
+ "answerable": {
+ "attempted": 30,
+ "completed": 30,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1666666666666667,
+ "r": 0.8333333333333334,
+ "recallAll": 0.8333333333333334,
+ "hit": 25,
+ "strictSuccessOverAttempts": 0.8333333333333334,
+ "latencyP50": 1.8002090000009048,
+ "latencyP95": 2.076791999999841,
+ "ingestionP50": 130.78970800000025
+ }
+ },
+ "temporal-reasoning": {
+ "grep": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 127,
+ "answerable": {
+ "attempted": 127,
+ "completed": 127,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.32755905511810973,
+ "r": 0.7692913385826772,
+ "recallAll": 0.6692913385826772,
+ "hit": 109,
+ "strictSuccessOverAttempts": 0.6692913385826772,
+ "latencyP50": 2.7004580000002534,
+ "latencyP95": 4.363542000000052,
+ "ingestionP50": 0.003750000000309228
+ },
+ "agentmemory-bm25-component": {
+ "n": 133,
+ "completed": 133,
+ "failed": 0,
+ "scored": 127,
+ "answerable": {
+ "attempted": 127,
+ "completed": 127,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.37795275590551125,
+ "r": 0.8803149606299214,
+ "recallAll": 0.7795275590551181,
+ "hit": 120,
+ "strictSuccessOverAttempts": 0.7795275590551181,
+ "latencyP50": 1.7312919999967562,
+ "latencyP95": 1.9771249999976135,
+ "ingestionP50": 129.1073749999996
+ }
+ },
+ "knowledge-update": {
+ "grep": {
+ "n": 78,
+ "completed": 78,
+ "failed": 0,
+ "scored": 72,
+ "answerable": {
+ "attempted": 72,
+ "completed": 72,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.3777777777777773,
+ "r": 0.9444444444444444,
+ "recallAll": 0.9166666666666666,
+ "hit": 70,
+ "strictSuccessOverAttempts": 0.9166666666666666,
+ "latencyP50": 2.0640830000002097,
+ "latencyP95": 3.4398329999999078,
+ "ingestionP50": 0.0018749999999272404
+ },
+ "agentmemory-bm25-component": {
+ "n": 78,
+ "completed": 78,
+ "failed": 0,
+ "scored": 72,
+ "answerable": {
+ "attempted": 72,
+ "completed": 72,
+ "failed": 0
+ },
+ "unanswerable": 6,
+ "unanswerableCompleted": 6,
+ "unanswerableFailed": 0,
+ "p": 0.394444444444444,
+ "r": 0.9861111111111112,
+ "recallAll": 0.9722222222222222,
+ "hit": 72,
+ "strictSuccessOverAttempts": 0.9722222222222222,
+ "latencyP50": 1.7199999999938882,
+ "latencyP95": 1.9851669999989099,
+ "ingestionP50": 127.13654200000019
+ }
+ },
+ "single-session-assistant": {
+ "grep": {
+ "n": 56,
+ "completed": 56,
+ "failed": 0,
+ "scored": 56,
+ "answerable": {
+ "attempted": 56,
+ "completed": 56,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.17857142857142852,
+ "r": 0.8928571428571429,
+ "recallAll": 0.8928571428571429,
+ "hit": 50,
+ "strictSuccessOverAttempts": 0.8928571428571429,
+ "latencyP50": 5.13412500000004,
+ "latencyP95": 7.704250000000229,
+ "ingestionP50": 0.0025000000000545697
+ },
+ "agentmemory-bm25-component": {
+ "n": 56,
+ "completed": 56,
+ "failed": 0,
+ "scored": 56,
+ "answerable": {
+ "attempted": 56,
+ "completed": 56,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1964285714285713,
+ "r": 0.9821428571428571,
+ "recallAll": 0.9821428571428571,
+ "hit": 55,
+ "strictSuccessOverAttempts": 0.9821428571428571,
+ "latencyP50": 1.8950000000040745,
+ "latencyP95": 2.241667000002053,
+ "ingestionP50": 127.22987499999726
+ }
+ }
+ },
+ "planned": 1000,
+ "completed": 1000,
+ "failed": 0,
+ "finishedAt": "2026-10-28T07:28:32.771Z",
+ "claim": "Development evaluation only. No competitor ranking or answer-accuracy claim."
+ },
+ "coding": {
+ "byAdapter": {
+ "grep": {
+ "n": 15,
+ "completed": 15,
+ "failed": 0,
+ "scored": 15,
+ "answerable": {
+ "attempted": 15,
+ "completed": 15,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.24000000000000002,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 15,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.03100000000000591,
+ "latencyP95": 0.19308300000000145,
+ "ingestionP50": 0.005083000000013271
+ },
+ "agentmemory-bm25-component": {
+ "n": 15,
+ "completed": 15,
+ "failed": 0,
+ "scored": 15,
+ "answerable": {
+ "attempted": 15,
+ "completed": 15,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.22666666666666666,
+ "r": 0.9333333333333333,
+ "recallAll": 0.9333333333333333,
+ "hit": 14,
+ "strictSuccessOverAttempts": 0.9333333333333333,
+ "latencyP50": 0.08995800000002419,
+ "latencyP95": 0.4154169999999908,
+ "ingestionP50": 1.278959000000043
+ }
+ },
+ "byType": {
+ "single-session-bug": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.19308300000000145,
+ "latencyP95": 0.19308300000000145,
+ "ingestionP50": 0.07908300000002555
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.4154169999999908,
+ "latencyP95": 0.4154169999999908,
+ "ingestionP50": 3.890874999999994
+ }
+ },
+ "single-session-infra": {
+ "grep": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02533299999998917,
+ "latencyP95": 0.03741600000000744,
+ "ingestionP50": 0.0025840000000130203
+ },
+ "agentmemory-bm25-component": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.1,
+ "r": 0.5,
+ "recallAll": 0.5,
+ "hit": 1,
+ "strictSuccessOverAttempts": 0.5,
+ "latencyP50": 0.08066700000000537,
+ "latencyP95": 0.14379100000002154,
+ "ingestionP50": 1.1475419999999872
+ }
+ },
+ "single-session-refactor": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.028250000000014097,
+ "latencyP95": 0.028250000000014097,
+ "ingestionP50": 0.007082999999965978
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.10045800000000327,
+ "latencyP95": 0.10045800000000327,
+ "ingestionP50": 1.338292000000024
+ }
+ },
+ "single-session-feature": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.024707999999975527,
+ "latencyP95": 0.024707999999975527,
+ "ingestionP50": 0.005375000000015007
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08070800000001555,
+ "latencyP95": 0.08070800000001555,
+ "ingestionP50": 1.6152920000000108
+ }
+ },
+ "single-session-test": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.03100000000000591,
+ "latencyP95": 0.03100000000000591,
+ "ingestionP50": 0.012625000000014097
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08995800000002419,
+ "latencyP95": 0.08995800000002419,
+ "ingestionP50": 1.11262499999998
+ }
+ },
+ "single-session-perf": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02583300000003419,
+ "latencyP95": 0.02583300000003419,
+ "ingestionP50": 0.006291999999973541
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.07287500000001046,
+ "latencyP95": 0.07287500000001046,
+ "ingestionP50": 1.278959000000043
+ }
+ },
+ "single-session-api": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.032708000000013726,
+ "latencyP95": 0.032708000000013726,
+ "ingestionP50": 0.0028750000000172804
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08158400000002075,
+ "latencyP95": 0.08158400000002075,
+ "ingestionP50": 1.1327079999999796
+ }
+ },
+ "single-session-db": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.035833000000025095,
+ "latencyP95": 0.035833000000025095,
+ "ingestionP50": 0.005083000000013271
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.13049999999998363,
+ "latencyP95": 0.13049999999998363,
+ "ingestionP50": 1.2863330000000133
+ }
+ },
+ "single-session-release": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.030915999999990618,
+ "latencyP95": 0.030915999999990618,
+ "ingestionP50": 0.002541000000007898
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.114291000000037,
+ "latencyP95": 0.114291000000037,
+ "ingestionP50": 1.3028339999999616
+ }
+ },
+ "multi-session-causal": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.04775000000000773,
+ "latencyP95": 0.04775000000000773,
+ "ingestionP50": 0.0028330000000096334
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.1052090000000021,
+ "latencyP95": 0.1052090000000021,
+ "ingestionP50": 1.1877920000000017
+ }
+ },
+ "preference": {
+ "grep": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.025458999999955267,
+ "latencyP95": 0.042124999999998636,
+ "ingestionP50": 0.0024999999999977263
+ },
+ "agentmemory-bm25-component": {
+ "n": 2,
+ "completed": 2,
+ "failed": 0,
+ "scored": 2,
+ "answerable": {
+ "attempted": 2,
+ "completed": 2,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.2,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 2,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.07791699999995672,
+ "latencyP95": 0.08508299999999736,
+ "ingestionP50": 1.0938330000000178
+ }
+ },
+ "multi-session-review": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.042709000000002106,
+ "latencyP95": 0.042709000000002106,
+ "ingestionP50": 0.0018750000000409273
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.12833399999999529,
+ "latencyP95": 0.12833399999999529,
+ "ingestionP50": 1.2722499999999854
+ }
+ },
+ "temporal": {
+ "grep": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.02400000000000091,
+ "latencyP95": 0.02400000000000091,
+ "ingestionP50": 0.004457999999999629
+ },
+ "agentmemory-bm25-component": {
+ "n": 1,
+ "completed": 1,
+ "failed": 0,
+ "scored": 1,
+ "answerable": {
+ "attempted": 1,
+ "completed": 1,
+ "failed": 0
+ },
+ "unanswerable": 0,
+ "unanswerableCompleted": 0,
+ "unanswerableFailed": 0,
+ "p": 0.4,
+ "r": 1,
+ "recallAll": 1,
+ "hit": 1,
+ "strictSuccessOverAttempts": 1,
+ "latencyP50": 0.08754099999998743,
+ "latencyP95": 0.08754099999998743,
+ "ingestionP50": 1.2965829999999983
+ }
+ }
+ },
+ "planned": 30,
+ "completed": 30,
+ "failed": 0,
+ "finishedAt": "2026-10-28T07:28:33.381Z",
+ "claim": "Development evaluation only. No competitor ranking or answer-accuracy claim."
+ }
+}
diff --git a/website/lib/generated-meta.json b/website/lib/generated-meta.json
index 0969a92d..64d8a8b6 100644
--- a/website/lib/generated-meta.json
+++ b/website/lib/generated-meta.json
@@ -1,15 +1,15 @@
{
- "version": "0.9.30",
+ "version": "0.9.31",
"mcpTools": 54,
"hooks": 12,
"restEndpoints": 146,
- "testsPassing": 3444,
- "stars": 29775,
- "forks": 2630,
- "contributors": 52,
- "releases": 52,
- "npmWeekly": 7096,
- "npmMonthly": 36978,
- "npmAllTime": 501164,
- "generatedAt": "2026-10-25T09:42:31.990Z"
+ "testCases": "4000+",
+ "stars": 29963,
+ "forks": 2646,
+ "contributors": 55,
+ "releases": 53,
+ "npmWeekly": 12481,
+ "npmMonthly": 42744,
+ "npmAllTime": 527228,
+ "generatedAt": "2026-10-28T08:19:41.991Z"
}
diff --git a/website/lib/llms.ts b/website/lib/llms.ts
index b59fc373..865ed299 100644
--- a/website/lib/llms.ts
+++ b/website/lib/llms.ts
@@ -6,17 +6,16 @@ import {
FAQ,
GATE_SCENARIOS,
INSTALL_CMD,
- LONGMEMEVAL,
REPO_URL,
SITE_URL,
- TOKENS,
src,
} from "./site";
import { MCP_JSON } from "./rest";
+import { EVALUATION, evaluationProse, percent } from "./bench";
import { getProjectMeta } from "./meta";
const SUMMARY =
- "agentmemory is open-source persistent memory for AI coding agents. Hooks capture what an agent does during a session, the local server stores and indexes it on the user's machine, and the next session starts with the relevant memories already in context. It runs locally with no account and no API key, and is licensed Apache-2.0.";
+ "agentmemory is open-source persistent memory for AI coding agents. Hooks capture what an agent does during a session, the server stores and indexes it on the user's machine by default, and the next session starts with the relevant memories already in context. It runs locally with no account and no API key, and is licensed Apache-2.0.";
export function llmsTxt(): string {
const m = getProjectMeta();
@@ -54,18 +53,17 @@ export function llmsTxt(): string {
export function llmsFullTxt(): string {
const m = getProjectMeta();
+ const summary = evaluationProse();
const plugins = AGENTS.filter((a) => a.support === "plugin");
const others = AGENTS.filter((a) => a.support !== "plugin");
const agentLine = (a: (typeof AGENTS)[number]) =>
`- ${a.name}: ${a.detail}${a.connect ? `. Wire it with \`agentmemory connect ${a.connect}\`` : ""}`;
- const h = LONGMEMEVAL.hybrid;
- const b = LONGMEMEVAL.bm25;
return [
"# agentmemory",
"",
SUMMARY,
"",
- `Version ${m.version}. ${m.mcpTools} MCP tools, ${m.hooks} capture hooks, ${m.restEndpoints} REST endpoints, ${m.testsPassing.toLocaleString("en-US")} tests. Website: ${SITE_URL}. Source: ${REPO_URL}.`,
+ `Version ${m.version}. ${m.mcpTools} MCP tools, ${m.hooks} capture hooks, ${m.restEndpoints} REST endpoints, ${m.testCases} test cases in the suite. Website: ${SITE_URL}. Source: ${REPO_URL}.`,
"",
"## Install",
"",
@@ -84,14 +82,14 @@ export function llmsFullTxt(): string {
"## How it works",
"",
"1. Capture. Hooks fire on every prompt and tool call. Each event is posted to the local server on port 3111 with a restart-safe id, or spooled to disk if the server is down.",
- "2. Store. Each event becomes an observation in the local data directory: the tool, the file, a bounded copy of the output, and the session it came from. The data directory is ~/Library/Application Support/agentmemory on macOS, $XDG_DATA_HOME/agentmemory or ~/.local/share/agentmemory on Linux, and %APPDATA%\\agentmemory on Windows; AGENTMEMORY_DATA_DIR overrides it.",
- "3. Index. Observations are indexed twice: BM25 terms for exact words and vectors for meaning. Local embeddings work with no API key.",
- "4. Consolidate. When the session ends, related observations collapse into memories that keep links back to every observation they came from.",
- "5. Recall. The next session opens with the relevant memories already in context.",
+ "2. Store. By default, each event becomes an observation in the local data directory: the tool, the file, a bounded copy of the output, and the session it came from. The data directory is ~/Library/Application Support/agentmemory on macOS, $XDG_DATA_HOME/agentmemory or ~/.local/share/agentmemory on Linux, and %APPDATA%\\agentmemory on Windows; AGENTMEMORY_DATA_DIR overrides it. A configured Redis backend stores records in Redis instead.",
+ "3. Index. BM25 indexes keywords by default. Optional local or remote embeddings add vector retrieval.",
+ "4. Summarize. Session summaries process new observations incrementally. Optional consolidation creates source-linked memories.",
+ "5. Recall. Connected integrations retrieve context when injection is enabled. MCP and REST also support explicit recall.",
"",
"## Agents",
"",
- `Native plugins (${plugins.length}), where hooks capture automatically:`,
+ `Native plugins (${plugins.length}): capture requires an installed plugin or hooks. Use --with-hooks where hook capture is opt-in.`,
"",
...plugins.map(agentLine),
"",
@@ -101,13 +99,12 @@ export function llmsFullTxt(): string {
"",
"## Benchmarks",
"",
- `LongMemEval-S, ${LONGMEMEVAL.questions} questions, retrieval recall (does a gold session appear in the top K results). Embedding model: ${LONGMEMEVAL.embedding}. No LLM in the loop. This measures retrieval, not end-to-end question answering accuracy.`,
- "",
- `- BM25 plus vectors: recall@5 ${h.r5}%, recall@10 ${h.r10}%, recall@20 ${h.r20}%, NDCG@10 ${h.ndcg10}%, MRR ${h.mrr}%`,
- `- BM25 only (keyless): recall@5 ${b.r5}%, recall@10 ${b.r10}%, recall@20 ${b.r20}%, NDCG@10 ${b.ndcg10}%, MRR ${b.mrr}%`,
- `- Source and reproduction steps: ${src.longmemeval}`,
- "",
- `Token cost: recalling from agentmemory used ${TOKENS.agentmemory.toLocaleString("en-US")} tokens where loading the full history used ${TOKENS.fullContext.toLocaleString("en-US")}, about ${TOKENS.savedPct}% fewer, measured on v0.6.0. Source: ${src.tokens}`,
+ summary.longmemeval,
+ `BM25 component strict recall-all@5 ${percent(EVALUATION.longmemeval.byAdapter["agentmemory-bm25-component"].recallAll)}; grep ${percent(EVALUATION.longmemeval.byAdapter.grep.recallAll)}. Unanswerable questions are excluded from retrieval metrics.`,
+ summary.multiSession,
+ "The metric requires every gold source among five distinct sessions. This component run excludes the daemon, embeddings, compression and answer generation. It is not a competitor ranking or release-to-release speed comparison.",
+ summary.coding,
+ `Methodology and receipts: ${src.longmemeval}`,
"",
"Release gate: every publish installs the packed package and runs these scenarios against it:",
"",
@@ -115,8 +112,8 @@ export function llmsFullTxt(): string {
"",
"## Privacy and security",
"",
- "- Memories and captured activity are stored in the local data directory on the user's machine (~/Library/Application Support/agentmemory on macOS, $XDG_DATA_HOME/agentmemory or ~/.local/share/agentmemory on Linux, %APPDATA%\\agentmemory on Windows). ~/.agentmemory holds configuration and the generated secret. There is no account, no hosted service and no telemetry.",
- "- Nothing leaves the machine unless the user configures an LLM or embedding provider; then that provider receives the text it needs. Local embeddings send nothing.",
+ "- Memories and captured activity are stored in the local data directory by default (~/Library/Application Support/agentmemory on macOS, $XDG_DATA_HOME/agentmemory or ~/.local/share/agentmemory on Linux, %APPDATA%\\agentmemory on Windows). ~/.agentmemory holds configuration and the generated secret. There is no account, no hosted service and no telemetry.",
+ "- Memory text stays local with local storage and keyless search or local models. Keyless search does not guarantee local storage: a configured remote Redis backend stores memory records remotely. Configured remote services receive the text they process. Installation, updates and first-use model downloads need network access.",
"- The REST API requires a bearer token by default. A secret is generated into ~/.agentmemory/secret on first start when AGENTMEMORY_SECRET is unset.",
"- The CLI's local REST and viewer listeners bind to 127.0.0.1 by default; the engine WebSocket is internal runtime traffic.",
`- Report vulnerabilities privately through GitHub Security Advisories, as described in ${REPO_URL}/blob/main/SECURITY.md`,
diff --git a/website/lib/meta.ts b/website/lib/meta.ts
index a362ae7f..36dead4b 100644
--- a/website/lib/meta.ts
+++ b/website/lib/meta.ts
@@ -7,7 +7,7 @@ export interface ProjectMeta {
mcpTools: number;
hooks: number;
restEndpoints: number;
- testsPassing: number;
+ testCases: string;
stars: number;
forks: number;
contributors: number;
diff --git a/website/lib/rest.ts b/website/lib/rest.ts
index ce0b7305..62c66817 100644
--- a/website/lib/rest.ts
+++ b/website/lib/rest.ts
@@ -27,18 +27,10 @@ export const FEATURED_LOGOS: Record<string, { src: string; wide?: boolean }> = {
"Product Hunt": { src: "https://cdn.simpleicons.org/producthunt" },
};
-export const INTERFACES = [
- {
- id: "viewer",
- label: "Viewer",
- port: ":3113",
- title: "The viewer ships with the server",
- body: "agentmemory starts a real-time viewer on port 3113 with no extra install. Watch observations arrive as hooks fire, replay any past session, browse memories and the knowledge graph, and check health on one page.",
- points: ["Live observation stream", "Session replay", "Memory and graph browser", "Health and status page"],
- img: "/demo.gif",
- width: 720,
- height: 405,
- alt: "agentmemory viewer showing a live stream of captured observations",
- launch: "open http://localhost:3113",
- },
-];
+export const VIEWER = {
+ features: ["Memories", "Session replay", "Lessons", "Knowledge graph", "Health"],
+ img: "/demo.gif",
+ width: 1080,
+ height: 610,
+ alt: "Recorded tour of the AgentMemory viewer with sample project data",
+};
diff --git a/website/lib/site.ts b/website/lib/site.ts
index a1f5ad72..93ee0c67 100644
--- a/website/lib/site.ts
+++ b/website/lib/site.ts
@@ -8,7 +8,9 @@ export const AGENT_INSTALL_PROMPT =
"Retrieve and follow the instructions at: https://raw.githubusercontent.com/rohitg00/agentmemory/main/INSTALL_FOR_AGENTS.md";
export const src = {
- longmemeval: `${REPO_URL}/blob/main/benchmark/LONGMEMEVAL.md`,
+ longmemeval: `${REPO_URL}/blob/main/docs/benchmarks/2026-10-28-retrieval-evaluation.md`,
+ evaluation: `${REPO_URL}/blob/main/eval/README.md`,
+ recall: `${REPO_URL}/blob/main/docs/recall.md`,
tokens: `${REPO_URL}/blob/main/benchmark/REAL-EMBEDDINGS.md`,
agentLife: `${REPO_URL}/blob/main/docs/benchmarks/2026-05-20-coding-agent-life-v1.md`,
gate: `${REPO_URL}/tree/main/scripts/release-gate`,
@@ -52,10 +54,11 @@ export const AGENTS: Agent[] = [
{ name: "Devin", logo: "/devin.png", href: "https://devin.ai", support: "plugin", detail: "6 hooks, skills, MCP", connect: "devin" },
{ name: "OpenClaw", logo: "https://github.com/openclaw.png", href: "https://github.com/openclaw/openclaw", support: "plugin", detail: "gateway plugin, MCP", connect: "openclaw" },
{ name: "Hermes", logo: "https://github.com/NousResearch.png", href: "https://github.com/NousResearch", support: "plugin", detail: "Python plugin, MCP", connect: "hermes" },
- { name: "pi", logo: "https://raw.githubusercontent.com/rohitg00/agentmemory/main/assets/agents/pi.svg", href: `${REPO_URL}/tree/main/integrations/pi`, support: "plugin", detail: "native plugin, MCP", connect: "pi" },
+ { name: "pi", logo: "https://raw.githubusercontent.com/rohitg00/agentmemory/main/assets/agents/pi.svg", href: `${REPO_URL}/tree/main/integrations/pi`, support: "plugin", detail: "installable package, capture, recall, skills", connect: "pi" },
+ { name: "Oh My Pi", logo: "https://raw.githubusercontent.com/rohitg00/agentmemory/main/assets/agents/pi.svg", href: `${REPO_URL}/tree/main/integrations/omp`, support: "plugin", detail: "capture, recall, skills, optional MCP", connect: "omp" },
{ name: "OpenHuman", logo: "https://raw.githubusercontent.com/tinyhumansai/openhuman/main/app/src-tauri/icons/128x128.png", href: "https://github.com/tinyhumansai/openhuman", support: "plugin", detail: "native Memory backend", connect: "openhuman" },
{ name: "Gemini CLI", logo: "https://github.com/google-gemini.png", href: "https://github.com/google-gemini/gemini-cli", support: "mcp", detail: "MCP server", connect: "gemini-cli" },
- { name: "Antigravity", logo: "https://svgl.app/library/antigravity.svg", href: "https://antigravity.google", support: "mcp", detail: "MCP, hooks", connect: "antigravity" },
+ { name: "Antigravity", logo: "https://svgl.app/library/antigravity.svg", href: "https://antigravity.google", support: "plugin", detail: "MCP, hooks", connect: "antigravity --with-hooks" },
{ name: "Claude Desktop", logo: "https://github.com/anthropics.png", href: "https://claude.ai/download", support: "mcp", detail: "MCP server" },
{ name: "Warp", logo: "https://svgl.app/library/warp.svg", href: "https://www.warp.dev", support: "mcp", detail: "MCP server", connect: "warp" },
{ name: "Zed", logo: "https://svgl.app/library/zed-logo_dark.svg", href: "https://zed.dev", support: "mcp", detail: "MCP server", connect: "zed" },
@@ -64,29 +67,13 @@ export const AGENTS: Agent[] = [
{ name: "Droid", logo: "https://www.factory.ai/favicon.svg", href: "https://docs.factory.ai/cli", support: "mcp", detail: "MCP server", connect: "droid" },
{ name: "Kiro", logo: "https://kiro.dev/favicon.ico", href: "https://kiro.dev", support: "mcp", detail: "MCP server", connect: "kiro" },
{ name: "Qwen Code", logo: "https://svgl.app/library/qwen_dark.svg", href: "https://github.com/QwenLM/qwen-code", support: "mcp", detail: "MCP server", connect: "qwen" },
- { name: "DeepSeek Harness", logo: "https://svgl.app/library/deepseek.svg", href: "https://github.com/deepseek-ai/deepseek-harness", support: "mcp", detail: "MCP server", connect: "dsh" },
+ { name: "DeepSeek Harness", logo: "https://svgl.app/library/deepseek.svg", href: "https://github.com/deepseek-ai/deepseek-harness", support: "plugin", detail: "native capture and recall, MCP", connect: "dsh --with-hooks" },
{ name: "Roo Code", logo: "https://github.com/RooCodeInc.png", href: "https://github.com/RooCodeInc/Roo-Code", support: "mcp", detail: "MCP server" },
{ name: "Kilo Code", logo: "https://github.com/Kilo-Org.png", href: "https://github.com/Kilo-Org/kilocode", support: "mcp", detail: "MCP server" },
{ name: "Goose", logo: "https://github.com/block.png", href: "https://github.com/block/goose", support: "mcp", detail: "MCP server" },
{ name: "Aider", logo: "https://github.com/Aider-AI.png", href: "https://github.com/Aider-AI/aider", support: "rest", detail: "REST API" },
];
-export const LONGMEMEVAL = {
- questions: 500,
- hybrid: { r5: 95.2, r10: 98.6, r20: 99.4, ndcg10: 87.9, mrr: 88.2 },
- bm25: { r5: 86.2, r10: 94.6, r20: 98.6, ndcg10: 73.0, mrr: 71.5 },
- byType: [
- { type: "knowledge-update", count: 78, r5: 98.7, r10: 100.0, h5: 77, h10: 78 },
- { type: "multi-session", count: 133, r5: 97.7, r10: 100.0, h5: 130, h10: 133 },
- { type: "single-session-assistant", count: 56, r5: 96.4, r10: 98.2, h5: 54, h10: 55 },
- { type: "temporal-reasoning", count: 133, r5: 95.5, r10: 97.7, h5: 127, h10: 130 },
- { type: "single-session-user", count: 70, r5: 90.0, r10: 97.1, h5: 63, h10: 68 },
- { type: "single-session-preference", count: 30, r5: 83.3, r10: 96.7, h5: 25, h10: 29 },
- ],
- embedding: "all-MiniLM-L6-v2, 384 dimensions, local",
-};
-
-export const TOKENS = { fullContext: 19462, agentmemory: 1571, savedPct: 92 };
export const GATE_SCENARIOS = [
{ id: "install", what: "The packed artifact installs and reports its identity" },
@@ -218,11 +205,11 @@ export const RESEARCH: Paper[] = [
export const FAQ = [
{
q: "What does agentmemory actually store?",
- a: "Observations from your agent's sessions: the prompts you typed, the tools it called and a bounded copy of their output, and the session summaries built from them. Consolidation turns related observations into memories, lessons and graph facts. Everything lives in a data folder on your machine, for example ~/Library/Application Support/agentmemory on macOS.",
+ a: "Observations from your agent's sessions: the prompts you typed, the tools it called and a bounded copy of their output, and the session summaries built from them. Consolidation turns related observations into memories, lessons and graph facts. By default, records live in a data folder on your machine, for example ~/Library/Application Support/agentmemory on macOS.",
},
{
q: "Does anything leave my machine?",
- a: "Not by default. With no provider configured, recall runs on BM25 keyword search and nothing is sent anywhere. If you add an LLM or an embedding provider, agentmemory sends that provider the text it needs. Local embeddings send nothing. There is no telemetry.",
+ a: "Memory text stays local when you use local storage and keyless search or local models. Keyless search does not guarantee local storage: a configured remote Redis backend stores memory records remotely. Installation, updates and the first local model download need network access. Configured remote LLM or embedding services receive the text they process. AgentMemory starts its runtime with usage telemetry disabled.",
},
{
q: "Do I need an API key?",
diff --git a/website/scripts/gen-meta.mjs b/website/scripts/gen-meta.mjs
index be00949a..9b82bd50 100644
--- a/website/scripts/gen-meta.mjs
+++ b/website/scripts/gen-meta.mjs
@@ -1,5 +1,5 @@
#!/usr/bin/env node
-import { readFileSync, readdirSync, writeFileSync, mkdirSync } from "node:fs";
+import { readFileSync, writeFileSync, mkdirSync } from "node:fs";
import { join, dirname } from "node:path";
import { fileURLToPath } from "node:url";
@@ -44,29 +44,6 @@ function countHookTypes(typesPath) {
.filter((s) => /^["'`]/.test(s)).length;
}
-function countTestCases(testDir) {
- let total = 0;
- let entries;
- try {
- entries = readdirSync(testDir, { withFileTypes: true });
- } catch {
- return 0;
- }
- for (const entry of entries) {
- const full = join(testDir, entry.name);
- if (entry.isDirectory()) {
- total += countTestCases(full);
- continue;
- }
- if (!/\.test\.[jt]sx?$/.test(entry.name)) continue;
- const txt = readFileSafe(full);
- if (!txt) continue;
- const m = txt.match(/(?:^|\s)(?:it|test)(?:\.\w+)?\s*\(/g);
- if (m) total += m.length;
- }
- return total;
-}
-
const pkg = safeReadJson(join(repoRoot, "package.json"));
const version = pkg?.version;
if (!version) {
@@ -85,7 +62,11 @@ const mcpTools = safeCountMatches(
/name:\s*"memory_/g,
);
const hooks = countHookTypes(join(repoRoot, "src", "types.ts"));
-const testsPassing = countTestCases(join(repoRoot, "test"));
+const synced = safeReadJson(join(repoRoot, "scripts/docs-sync.state.json"));
+const testCases = synced?.facts?.tests?.[0];
+if (synced?.version !== version || !testCases || !mcpTools || !hooks || !restEndpoints) {
+ throw new Error("gen-meta: source metadata is missing or stale; run npm run docs:sync from the repository root");
+}
const REPO = "rohitg00/agentmemory";
const NPM_PKG = "@agentmemory/agentmemory";
@@ -183,16 +164,28 @@ for (const key of Object.keys(STAT_FALLBACK)) {
const meta = {
version,
- mcpTools: mcpTools || 45,
- hooks: hooks || 12,
- restEndpoints: restEndpoints || 107,
- testsPassing: testsPassing || 794,
+ mcpTools,
+ hooks,
+ restEndpoints,
+ testCases,
...stats,
generatedAt: new Date().toISOString(),
};
writeFileSync(outPath, JSON.stringify(meta, null, 2) + "\n", "utf-8");
+const evaluation = {};
+for (const suite of ["longmemeval", "coding"]) {
+ const dir = join(repoRoot, "docs/benchmarks/receipts/2026-10-28", suite);
+ const summary = safeReadJson(join(dir, "summary.json"));
+ const manifest = safeReadJson(join(dir, "manifest.json"));
+ if (!summary || manifest?.k !== 5 || summary.completed + summary.failed !== summary.planned) {
+ throw new Error(`gen-meta: missing or incomplete ${suite} evaluation receipts`);
+ }
+ evaluation[suite] = summary;
+}
+writeFileSync(join(outDir, "generated-evaluation.json"), JSON.stringify(evaluation, null, 2) + "\n");
+
function parseChangelog(text) {
const releases = [];
const re = /^## \[(\d+\.\d+\.\d+)\][^\n]*?(\d{4}-\d{2}-\d{2})?\s*$/gm;
@@ -213,5 +206,5 @@ const changelog = parseChangelog(readFileSafe(join(repoRoot, "CHANGELOG.md"))).s
writeFileSync(join(outDir, "generated-changelog.json"), JSON.stringify(changelog, null, 2) + "\n", "utf-8");
console.log(
- `[gen-meta] wrote ${outPath}: v${meta.version}, ${meta.mcpTools} tools, ${meta.hooks} hooks, ${meta.restEndpoints} endpoints, ${meta.testsPassing} tests, ${meta.stars} stars, ${meta.npmMonthly} npm/month, ${changelog.length} changelog entries`,
+ `[gen-meta] wrote ${outPath}: v${meta.version}, ${meta.mcpTools} tools, ${meta.hooks} hooks, ${meta.restEndpoints} endpoints, ${meta.testCases} test cases, ${meta.stars} stars, ${meta.npmMonthly} npm/month, ${changelog.length} changelog entries`,
);