Turbocharge Claude Code, Cursor, Codex, Gemini & every coding agent: faster, cheaper, with contextual understanding specific to your codebase.
TypeScript ai-agentsanthropicclaude-codeclicode-graphcodexcontext-engineeringcursordeveloper-toolsgeminiknowledge-graphllmmcpmcp-serveropen-sourceopenaitree-sitter
Reposition around repo-native markdown graph: rewrite README + web landing page
anirudhkumar-nanonetsf9fd1f55b026
diff --git a/README.md b/README.mdindex 149c3254..3a6ada64 100644--- a/README.md+++ b/README.md@@ -1,420 +1,34 @@# Context Graph Engine-> Turn any docs into a structured **context graph** that AI agents read from before doing work — and contribute their learnings back to, so the graph gets smarter every time it's used.+**31% fewer tool calls, 23% less cost, 17% lower latency.** That's what your agent gets from reading the repo's context graph before it starts working.-Dumping docs into a vector store gives you a bag of chunks. It doesn't give you *structure* an agent can reason over, and there's no clean way for an agent to send back what it learned mid-task. The Context Graph Engine fixes both:--- **Ingest** any docs → a graph of **entities** (concepts, systems, people, APIs…) connected by typed **relationships**, grounded in the original source passages.-- **Read** the graph before a task → get a compact, structured context bundle (entities + relationships + supporting sources), ready to drop into a prompt.-- **Contribute** learnings back → new facts are deduplicated and merged into existing knowledge; repeated observations **reinforce** confidence. The graph compounds over time.--Access it four ways: a **web UI** (your team's knowledge as a living, citable graph — see below), a **library**, a **CLI**, or an **MCP server** that any agent (Claude Code, Cursor, …) can plug into.--> **Benchmark:** on a real 344-file repo, giving the agent the graph up front cut tool calls **31%**, cost **23%**, and latency **17%** vs. exploring cold — same model, same tools, only difference is the graph bundle.-------## Install--**One line, runs locally, no API keys required:**--```bash-curl -fsSL https://raw.githubusercontent.com/NanoNets/context-graph-engine/main/install.sh | sh-```--This puts `context-graph`, `context-graph-mcp`, and `context-graph-web` on your PATH. Requires git and Node ≥ 20.--Or straight from GitHub with npm (also works as a library / dev dependency):--```bash-npm install -g github:NanoNets/context-graph-engine-```--## Try it in 60 seconds — the web UI--```bash-context-graph-web-# → http://localhost:4680-```--Open the page, drop a folder of `.pdf` / `.md` / `.txt` files on the **“Add documents”** dropzone, and watch it become a knowledge graph. Then ask questions: answers come back with numbered citations, and every fact wears a **provenance receipt** — how many times it's been observed, how confident the graph is, and which sources say so. Use **“Add a fact”** to record something the docs don't capture, and watch the graph reinforce in real time.--Everything runs on your machine (the server binds to localhost only). Set `OPENROUTER_API_KEY` to upgrade to LLM-written answers and cloud extraction; without it, retrieval and the graph stay fully local.--Running it for your whole team? One `docker compose up` gives everyone the same graph at one URL — see [Team sharing](#team-sharing).--### Or run everything at once — `context-graph serve`--```bash-context-graph serve-# → web UI http://localhost:4680-# MCP (HTTP) http://localhost:4680/mcp-```--One process runs the web UI, the MCP server (over HTTP), and the folder watcher on a single port over **one shared engine** — the supported way to run the whole thing at once, with no separate `watch` daemon to babysit. Open the UI, click **Connect a folder**, tick **Keep watching**, and that folder is ingested and watched live; it's remembered, so the next `serve` resumes it. Prefer the command line? Pass folders directly: `context-graph serve ./sample-docs`.--### Runs locally by default — keys are optional--Out of the box the engine needs **no accounts and no cloud calls**:--- **Embeddings** run **in-process** via a small ONNX model (transformers.js) — downloaded and cached on first use.-- **Extraction** uses a local **[Ollama](https://ollama.com)** model. Install Ollama and pull one:- ```bash- ollama pull llama3.2- ```--Want higher-quality cloud models instead? Set a key and the engine uses it **automatically** (local remains the fallback):--```bash-export OPENROUTER_API_KEY=sk-or-... # cloud extraction via OpenRouter-export OPENAI_API_KEY=sk-... # cloud embeddings (optional; local otherwise)-```--Force local even when keys are present with `CONTEXT_GRAPH_LOCAL=1`.-------## Quickstart (library)--```ts-import { ContextGraphEngine } from "context-graph-engine";--const engine = new ContextGraphEngine(); // persists to ./.context-graph/graph.db--// 1. Ingest docs — builds the graph.-await engine.ingest(myMarkdownDocs, { title: "Onboarding" });-await engine.ingestFile("./docs/architecture.md");--// 2. Read before doing work.-const ctx = await engine.read("how does authentication work?");-console.log(ctx.prompt); // inject this string into your agent-console.log(ctx.nodes); // structured entities-console.log(ctx.edges); // structured relationships--// 3. Contribute what an agent learned.-await engine.contribute(- "The auth service rotates tokens every 15 minutes.",- { agentId: "code-reviewer" },-);-```--Run the full working demo (set `OPENROUTER_API_KEY`, or run a local Ollama):--```bash-npm run example-```-------## CLI--```bash-# Ingest documents (Markdown, text, and PDFs — PDFs are parsed automatically)-context-graph ingest ./docs/*.md ./handbook.pdf-context-graph ingest-dir ./docs # a whole folder, recursively-echo "some notes" | context-graph ingest-text --title "Notes"--# Keep the graph evolving as a folder changes (see "Auto-watch")-context-graph watch ./docs # register + watch; re-run with no args to resume-context-graph ingest-dir ./docs --watch # ingest now, register for watching-context-graph watch-status # list registered folders-context-graph unwatch ./docs--# Ingest a code repository — as summaries, not raw code (see "Code repositories")-context-graph repo .--# Ingest GitHub pull requests (title, description, review discussion) via the gh CLI-context-graph ingest-prs --state merged --limit 50--# Wire the graph into Claude Code (see "Claude Code hooks")-context-graph install-hooks--# Visualize the graph (interactive HTML; also --format json|mermaid)-context-graph export --out graph.html && open graph.html--# Read the graph-context-graph query "how does billing retry failed charges?"-context-graph query "billing" --json # full structured bundle--# Contribute a learning-context-graph contribute "Dunning worker retries 3x then marks past_due" --agent me--# Inspect-context-graph stats--# Point at a specific graph file-context-graph --db ./team-graph.db query "auth"--# Team sharing — git mode (see "Team sharing" below)-context-graph sync # pull teammates' graph.jsonl, re-merge, write it back-context-graph push # write the graph to a committable graph.jsonl-context-graph pull # import + re-merge a teammate's graph.jsonl-```-------## MCP server--Expose the graph to any MCP client so agents can `context_read` and `context_contribute` directly — the same "shared skill / durable memory" model as tools like ByteRover.--Add to your Claude Code / MCP client config:--```json-{- "mcpServers": {- "context-graph": {- "command": "context-graph-mcp",- "env": {- "CONTEXT_GRAPH_DB": "/absolute/path/to/team-graph.db"- }- }- }-}-```--Or, if you're running `context-graph serve`, point any MCP client at its HTTP endpoint instead of spawning a process:--```json-{- "mcpServers": {- "context-graph": { "type": "http", "url": "http://localhost:4680/mcp" }- }-}-```--When the server is exposed beyond localhost, add `"headers": { "Authorization": "Bearer <access-token>" }` — the same token the web UI prints.--The API-key env vars are optional — add `OPENROUTER_API_KEY` (and optionally `OPENAI_API_KEY` for embeddings) only if you want cloud models instead of the local defaults.--Tools exposed:--| Tool | What it does |-|------|--------------|-| `context_read` | Read relevant context for a query before working |-| `context_contribute` | Write a learning back into the shared graph |-| `context_ingest` | Ingest a document (raw text) into the graph |-| `context_ingest_file` | Ingest files from disk, including **PDFs** (parsed automatically) |-| `context_ingest_dir` | Ingest a whole **directory** of docs (PDF/MD/TXT by default; pass `extensions` to widen), recursively |-| `context_ingest_repo` | Ingest a **code repository** as per-file prose summaries (never raw code); incremental across runs |-| `context_watch_dir` | Connect a folder for **auto-watching**: ingest it now and re-ingest files as they change (see "Auto-watch") |-| `context_unwatch_dir` | Disconnect a watched folder (its knowledge stays in the graph) |-| `context_watch_status` | List connected folders and the ingest queue |-| `context_export` | Write the graph to an interactive **HTML** visualization |-| `context_sync` | Team sharing (git mode): import + re-merge the shared `graph.jsonl`, then write it back |-| `context_stats` | Report how much the graph currently holds |--A natural agent workflow: **read context → do the task → contribute what you learned.**-------## Code repositories--`context-graph repo <dir>` (or the `context_ingest_repo` MCP tool) ingests a codebase — but deliberately **never feeds raw source code to the entity extractor**. Running LLM extraction over thousands of code chunks is the most expensive way to compute what `grep` and tree-sitter give agents for free, and it fills the graph with noisy, duplicated symbol entities. (Every serious code-context system — aider, Cursor, Sourcegraph, Claude Code — derives code *structure* statically, and uses LLMs only for prose.)--Instead, repo ingestion:--1. Walks the repo for code files (`.ts`, `.py`, `.go`, … — `node_modules`/`dist`/etc. and files over 1 MB are skipped).-2. Writes **one LLM prose summary per file**: purpose, key exports, dependencies, design decisions.-3. Groups summaries into one document per top-level directory and ingests those through the normal pipeline — so the graph gains *module-level* entities ("the chunker is paragraph-aware", "the CLI requires the GitHub CLI for PR ingestion") with provenance, not ten thousand function names.--It is **incremental**: summaries are cached by content hash in `repo-summaries.json` next to the db, so re-running after edits only re-summarizes changed files, and a changed file's module document *replaces* its predecessor rather than piling up stale duplicates. An unchanged repo re-ingests with **zero** LLM calls.--The same replace-on-change behavior now applies to every file ingest: re-ingesting a modified doc replaces the old version instead of leaving both in the graph.--For a ~30-file repo expect ~30 small summary calls on first run; the graph stays module-scale (a handful of documents) rather than chunk-soup-scale. Agents should still read raw code with their own tools — the graph holds what code can't say: intent, decisions, and gotchas.--## Auto-watch — an evolving graph--Connect a folder once and the graph tracks it: files you add or edit are re-ingested automatically, so the graph evolves with your notes instead of freezing at ingest time.--```bash-context-graph watch ~/notes # catch up on the folder, then watch it live-# … later, in a fresh shell — no args needed, registered folders resume:-context-graph watch-```--How it behaves:--- **Debounced and cheap.** Saves are coalesced (~1.5 s quiet time per file) and unchanged files are skipped by content hash, so a daemon restart over a big unchanged folder costs zero LLM calls. Edited files *replace* their old document rather than piling up copies.-- **Append-only.** Deleting a file never removes knowledge already learned from it — the graph is a memory, not a mirror. The deletion is logged and nothing else happens.-- **Same rules as `ingest-dir`.** Default extensions `.pdf .md .markdown .txt` (override with `--ext`), dot-dirs / `node_modules` / files over 1 MB are ignored.-- **Registered folders** live in `watched-dirs.json` next to the db (machine-local — it never rides the git-synced graph). `ingest-dir --watch` registers while ingesting; `watch-status` / `unwatch` manage the list.-- **One watcher per db.** Run a single `watch` daemon per graph; MCP sessions and the web UI can read concurrently while it runs.--Elsewhere: the `context_watch_dir` MCP tool connects a folder for the life of the agent session (set `CONTEXT_GRAPH_AUTOWATCH=1` on the MCP server to auto-resume registered folders), and `CONTEXT_GRAPH_WATCH=1` does the same for `context-graph-web` — the UI picks up growth live.--## Claude Code hooks--`context-graph install-hooks` wires the graph into Claude Code via `.claude/settings.json` (merges with what's already there):--- **SessionStart** — every new session begins with relevant graph context (conventions, decisions, gotchas) auto-injected, plus a reminder that `context_contribute` exists.-- **Stop** — when the agent finishes, it's nudged once to record any durable learnings from the session before ending. Pass `--no-stop` to skip this hook if you find the nudge too chatty.--Both hooks stay silent when there's no graph yet, so installing them in a fresh project is harmless. Combined with the MCP server this closes the loop: sessions **start** with team knowledge and **end** by adding to it.--## GitHub pull requests--PR descriptions and review threads are where decisions actually get written down. `context-graph ingest-prs` pulls them into the graph via the [GitHub CLI](https://cli.github.com) (no extra auth needed if `gh` is logged in):--```bash-context-graph ingest-prs # merged PRs of the current repo (default 50)-context-graph ingest-prs -R owner/name -n 100 --state all-```--Each PR becomes one document (title, description, review discussion) with source `pr:owner/name#123`. Re-runs are cheap: unchanged PRs are skipped by hash, edited ones replace their old version.-------## Team sharing--A solo graph is one SQLite file. To share one across a team, pick a mode. All three keep the local-first default intact (team sync is **opt-in**).--### The easy mode — one shared server (recommended)--This is how nearly every self-hosted team tool works (Vaultwarden, Outline, BookStack): one instance, one URL, everyone's browser. No sync to configure, contributions land in the same graph instantly, and there is exactly one copy of the truth.--```bash-docker compose up -d-cat ./data/share-link.txt # the share link, with access token-```--The graph persists to `./data/graph.db`. Ingest documents two ways: click **Choose a folder…** in the UI to upload a folder straight from your computer (works for remote teammates), or drop files into `./docs` on the server and ingest the mounted `/docs` path. When exposed beyond localhost the API requires an access token. It's generated once, saved to `./data/share-link.txt` (and printed to `docker compose logs`), and **stays stable across restarts** so shared links keep working. Pin your own with `CONTEXT_GRAPH_WEB_TOKEN`.--No server to put it on? Run it on any machine and share it over [Tailscale](https://tailscale.com) without opening a port to the internet:--```bash-context-graph-web &-tailscale serve localhost:4680 # private HTTPS URL, visible only to your tailnet-```--### Git mode — sync separate graphs, no server at all--Prefer no shared server? Each teammate keeps their own local graph and git moves the knowledge around. This relies on the graph's **conflict-free merge**: observations are a grow-only counter and free-text fields are last-writer-wins, so re-importing the same facts never double-counts and merge order never changes the result.--Commit the graph to your repo as a human-diffable file and let git move it around.--```bash-# You: capture the merged graph and commit it-context-graph sync-git add .context-graph/graph.jsonl && git commit -m "sync context graph" && git push--# A teammate: get the latest and converge their local graph-git pull-context-graph sync-```--`sync` imports the committed `.context-graph/graph.jsonl`, re-merges it into the local graph, and writes the merged result back. It's idempotent (safe to run repeatedly) and commutative (order-independent), so two teammates' additions always converge. The repo's `.gitignore` keeps the local SQLite replica private while tracking `graph.jsonl`. Best for teams already living in git who want no servers and offline-friendly sharing; the only friction is the occasional merge conflict on the file, resolved by re-running `sync`.--> Both modes assume one embedding model per graph (see the note under *How it works*). Imports across mismatched embedding dimensions keep the facts but drop the incompatible vectors, and warn you to re-ingest sources to re-embed.-------## How the graph gets smarter over time--Every entity and relationship carries a `confidence` score and an `observations` count.--- **Dedup on write.** When new knowledge comes in, each entity is matched against existing nodes by exact name/alias **and** embedding similarity (cosine ≥ `mergeThreshold`). A match is *merged*, not duplicated.-- **Reinforcement.** Each observation is recorded under the id of the document/contribution that made it, so `observations` is a grow-only counter and `confidence` is *derived* from it (seen-often facts rise toward 1.0; one-offs stay low). Because the counter merges by taking the max per source, re-importing the same records is a no-op — reinforcement survives team sync without double-counting.-- **Provenance.** Every node/edge tracks which documents and agents contributed to it.-- **Grounding.** Source passages are kept as retrievable evidence and returned alongside the structured graph.--This is what makes contributions compound: the tenth agent to confirm a fact strengthens it rather than cluttering the graph.-----+The graph is a folder of plain markdown files in the repo — one per system, API, or concept, linked to each other. It stays in sync with the code through git, so anyone who clones the repo has it, no setup.## How it works-```- ingest / contribute read- ┌──────────────────────────┐ ┌────────────────────────┐- │ text → chunk → embed │ │ query → embed │- │ → extract (LLM) │ │ → semantic match nodes │- │ → MERGE (dedup + │───────────▶│ → expand 1 hop (edges) │- │ reinforce) │ graph │ → gather source chunks │- └──────────────────────────┘ (SQLite) │ → render context bundle│- └────────────────────────┘-```--- **Storage** — SQLite (via `better-sqlite3`) by default, or swap in your own `GraphStore`.-- **Extraction** — local **Ollama** (`llama3.2` by default) using structured JSON output; automatically upgrades to **OpenRouter** (`openai/gpt-4o-mini` by default, any tool-calling model) if `OPENROUTER_API_KEY` is set.-- **Embeddings** — local **in-process** model (`Xenova/all-MiniLM-L6-v2`, 384-dim) by default; automatically upgrades to **OpenAI** (`text-embedding-3-small`) if `OPENAI_API_KEY` is set.-- **Retrieval** — semantic match over entities + one-hop graph expansion + supporting source passages.--> Note: embeddings from different models aren't comparable, so pick local **or** cloud for a given database and stick with it — don't switch providers mid-graph.-------## Configuration--Everything is configurable via constructor options, environment variables, or defaults (in that order):--```ts-new ContextGraphEngine({- dbPath: "./.context-graph/graph.db", // CONTEXT_GRAPH_DB- // Cloud (used automatically when set):- openrouterApiKey: process.env.OPENROUTER_API_KEY,- openrouterModel: "openai/gpt-4o-mini", // CONTEXT_GRAPH_OPENROUTER_MODEL- openaiApiKey: process.env.OPENAI_API_KEY, // optional, embeddings only- embeddingModel: "text-embedding-3-small", // CONTEXT_GRAPH_EMBEDDING_MODEL- // Local (the zero-key fallback):- forceLocal: false, // CONTEXT_GRAPH_LOCAL=1 forces local- localEmbeddingModel: "Xenova/all-MiniLM-L6-v2",// CONTEXT_GRAPH_LOCAL_EMBEDDING_MODEL- ollamaModel: "llama3.2", // CONTEXT_GRAPH_OLLAMA_MODEL- ollamaBaseUrl: "http://localhost:11434", // CONTEXT_GRAPH_OLLAMA_URL- chunkSize: 1200,- chunkOverlap: 200,- mergeThreshold: 0.86, // higher = stricter dedup-});-```+- Edit code, regenerate the graph, commit both together.+- Someone clones the repo — they've got the graph, nothing to set up.+- Their agent reads it before doing work.-### Bring your own components+No database, no server, no embeddings. The graph is just files in git, so git is the sync. And when a change moves something around, you see it in the graph diff right next to the code, in the same PR.-The engine depends only on small interfaces, so you can replace any piece:+## Setup-```ts-import { ContextGraphEngine, type Embedder, type Extractor, type GraphStore, type Summarizer } from "context-graph-engine";--new ContextGraphEngine({- store: myCustomStore, // e.g. Postgres/pgvector- embedder: myCustomEmbedder, // any embedding model- extractor: myCustomExtractor, // any LLM, or a rules-based extractor- summarizer: myCustomSummarizer // code-file summaries for ingestRepo-});-```--This is exactly how the test suite runs the whole pipeline offline with fake providers.-------## API reference (essentials)+```bash+# in your repo+context-graph init # builds .context/ from your code+git add .context && git commit-```ts-class ContextGraphEngine {- ingest(text: string, opts?: { title?; source? }): Promise<IngestResult>;- ingestFile(path: string, opts?): Promise<IngestResult>; // .pdf parsed automatically- ingestDir(dir: string, opts?): Promise<IngestResult[]>; // recursive; PDF/MD/TXT- ingestRepo(dir: string, opts?): Promise<RepoIngestResult>; // code → per-file summaries, incremental- read(query: string, opts?: { maxNodes?; maxChunks?; expand? }): Promise<ContextBundle>;- contribute(learning: string, opts?: { agentId?; source? }): Promise<ContributeResult>;- exportGraph(): GraphExport; // -> toHtml / toMermaid- stats(): GraphStats;- close(): void;-}+# keep it honest — fails CI if the graph is stale+context-graph check```-A `ContextBundle` contains `{ query, nodes, edges, chunks, prompt }` — use `prompt` for a ready-made context block, or the structured arrays to build your own.-----+That's it. Anyone who clones the repo gets `.context/` for free. Point your agent at it and it reads the graph before working.-## Development+## The proof-```bash-npm install-npm run build # compile to dist/-npm run cli -- stats # run the CLI from source-npm run mcp # run the MCP server from source-```+Same model, same tools, same tasks. The only difference is whether the agent got the graph first. On a real 344-file repo it cut tool calls 31%, cost 23%, and latency 17%. The graph isn't magic — it's just that reading a few linked files beats rediscovering the codebase from scratch every single time.-## License+## Runs locally-MIT+Everything runs on your machine. No accounts, no cloud calls, no keys. Building the graph uses a local model (Ollama); if you'd rather use a cloud model for better summaries, set a key and it'll use it. Local stays the default.diff --git a/src/web/ui.html b/src/web/ui.htmlindex e50bf3ee..80bbec8b 100644--- a/src/web/ui.html+++ b/src/web/ui.html@@ -3,7 +3,7 @@<head><meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />-<title>Context Graph — your team's living knowledge ledger</title>+<title>Context Graph — read the graph before you work</title><link rel="icon" href="data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 32 32'%3E%3Ccircle cx='9' cy='9' r='4' fill='%232E6B4E'/%3E%3Ccircle cx='24' cy='13' r='3' fill='%231C2422'/%3E%3Ccircle cx='14' cy='24' r='3' fill='%231C2422'/%3E%3Cpath d='M12 11l9 2M11 12l3 9M23 16l-8 7' stroke='%239AA79F' stroke-width='1.5'/%3E%3C/svg%3E" /><style>:root {@@ -16,1479 +16,179 @@--verdant: #2e6b4e;--verdant-deep: #1f4f38;--verdant-soft: #e2ede4;- --amber: #a9762b;- --amber-soft: #f3ead8;--serif: "Charter", "Iowan Old Style", "Palatino", Georgia, serif;--sans: system-ui, -apple-system, "Segoe UI", sans-serif;--mono: ui-monospace, "SF Mono", SFMono-Regular, Menlo, Consolas, monospace;- /* spacing — 4px base */--s1: 4px; --s2: 8px; --s3: 12px; --s4: 16px;--s5: 24px; --s6: 32px; --s7: 48px; --s8: 64px;- /* type roles */- --fs-label: 11px;- --fs-meta: 12px;- --fs-ui: 13.5px;- --fs-body: 15px;- --fs-read: 17px;- --fs-title: 20px;- --fs-display: 26px;- /* shape & elevation */--r-sm: 4px; --r-md: 6px; --r-lg: 10px;--shadow-1: 0 1px 3px rgba(28,36,34,.06);- --shadow-2: 0 6px 24px rgba(28,36,34,.10);- /* motion */- --dur-fast: 160ms; --dur-med: 320ms; --dur-slow: 600ms;--ease-out: cubic-bezier(.22,.7,.3,1);}* { box-sizing: border-box; margin: 0; padding: 0; }- html, body { height: 100%; }body {font-family: var(--sans);background: var(--paper);color: var(--ink);- font-size: var(--fs-body);+ font-size: 15px;line-height: 1.5;+ -webkit-font-smoothing: antialiased;}- button, input, textarea { font: inherit; color: inherit; }- :focus-visible { outline: 2px solid var(--verdant); outline-offset: 2px; }+ @keyframes rise { from { opacity: 0; transform: translateY(8px); } to { opacity: 1; transform: none; } }+ .rise { animation: rise 480ms var(--ease-out) both; animation-delay: calc(var(--i, 0) * 70ms); }+ @media (prefers-reduced-motion: reduce) { .rise { animation: none; } }- /* ---------- motion primitives ---------- */- @keyframes rise { from { opacity: 0; transform: translateY(6px); } to { opacity: 1; transform: none; } }- .rise { animation: rise var(--dur-med) var(--ease-out) both; animation-delay: calc(var(--i, 0) * 45ms); }- @keyframes shimmer { to { background-position: -200% 0; } }- @keyframes spin { to { transform: rotate(360deg); } }+ code { font-family: var(--mono); }- /* ---------- app state ---------- */- /* Until boot resolves the state, both halves stay unmounted — no flash. */- #onboarding { display: none; }- body.is-empty #onboarding { display: block; }- #ask-section, #library { display: none; }- body.is-ready #ask-section { display: block; }- body.is-ready #library { display: block; }-- /* ---------- header ---------- */+ /* ---------- masthead ---------- */.masthead {display: flex; align-items: baseline; gap: 14px; flex-wrap: wrap;padding: 14px 22px 12px;border-bottom: 1px solid var(--line);background: var(--panel);}- .wordmark { font-family: var(--sans); font-weight: 700; font-size: 15px; letter-spacing: .14em; text-transform: uppercase; }+ .wordmark { font-weight: 700; font-size: 15px; letter-spacing: .14em; text-transform: uppercase; }.wordmark .dot { color: var(--verdant); }.tagline { font-family: var(--serif); font-style: italic; color: var(--ink-soft); font-size: 14.5px; }- .masthead-right { margin-left: auto; display: flex; align-items: baseline; gap: var(--s2); }- .stat { font-family: var(--mono); font-size: var(--fs-meta); color: var(--ink-soft); }- .stat b { color: var(--ink); font-size: 14px; font-weight: 600; }- .stat-sep { color: var(--ink-faint); font-family: var(--mono); font-size: var(--fs-meta); }- .provider-badge {- font-family: var(--mono); font-size: var(--fs-label); padding: 3px 9px;- border: 1px solid var(--line); border-radius: 999px; color: var(--ink-soft);- background: var(--paper); margin-left: var(--s3);- }- .provider-badge.cloud { border-color: var(--verdant); color: var(--verdant-deep); background: var(--verdant-soft); }- .provider-badge:not(.cloud) { cursor: pointer; }-- /* ---------- layout ---------- */- .frame {- display: grid;- grid-template-columns: minmax(400px, 500px) 1fr;- height: calc(100vh - 49px);- }- .desk {- overflow-y: auto;- padding: var(--s5) var(--s5) var(--s8);- border-right: 1px solid var(--line);- }- .desk > section { margin-bottom: var(--s6); }- .atlas { position: relative; background: var(--panel); }- #graph { position: absolute; inset: 0; }-- /* ---------- onboarding hero ---------- */- .hero h1 { font-family: var(--serif); font-size: var(--fs-display); font-weight: 600; margin: var(--s4) 0 var(--s3); }- .hero-sub { color: var(--ink-soft); font-size: var(--fs-body); line-height: 1.55; margin-bottom: var(--s5); max-width: 46ch; }- .steps { list-style: none; margin-bottom: var(--s5); display: flex; flex-direction: column; gap: var(--s2); }- .steps li { display: flex; gap: var(--s3); align-items: center; font-size: var(--fs-ui); color: var(--ink-soft); }- .steps li b { color: var(--ink); }- .steps .num {- font-family: var(--mono); font-size: var(--fs-label); font-weight: 700;- color: var(--verdant-deep); background: var(--verdant-soft); border-radius: 50%;- width: 20px; height: 20px; flex: none;- display: inline-flex; align-items: center; justify-content: center;- }- .dropzone {- display: flex; flex-direction: column; gap: var(--s1); align-items: center; justify-content: center;- width: 100%; padding: var(--s6) var(--s4);- border: 1.5px dashed var(--ink-faint); border-radius: var(--r-lg);- background: var(--panel); cursor: pointer; text-align: center;- transition: border-color var(--dur-fast), background var(--dur-fast), transform var(--dur-fast);- }- .dropzone:hover, .dropzone.drag { border-color: var(--verdant); background: var(--verdant-soft); }- .dropzone.drag { transform: scale(1.01); }- .dropzone:disabled { opacity: .55; cursor: default; }- .dz-main { font-family: var(--serif); font-size: var(--fs-title); color: var(--ink); }- .dz-sub { font-family: var(--mono); font-size: var(--fs-meta); color: var(--ink-soft); }- /* compact variant inside the library drawer */- #ingest-panel .dropzone { padding: var(--s4); }- #ingest-panel .dz-main { font-size: var(--fs-body); }-- #ingest-controls .picker-row { margin-top: var(--s3); }-- /* ---------- ingest progress ---------- */- #ingest-progress { margin-top: var(--s3); }- .progress-track { height: 3px; background: var(--line); border-radius: 2px; overflow: hidden; }- .progress-track i { display: block; height: 100%; width: 0; background: var(--verdant); transition: width var(--dur-med) var(--ease-out); }- #progress-rows { margin-top: var(--s2); font-family: var(--mono); font-size: var(--fs-meta); color: var(--ink-soft); }- .progress-row { display: flex; gap: var(--s2); align-items: center; padding: 2px 0; min-width: 0; }- .progress-row > span:last-child { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }- .progress-row .icon { width: 12px; flex: none; text-align: center; }- .progress-row.ok .icon { color: var(--verdant); }- .progress-row.err { color: #a03b2e; }- .spinner {- display: inline-block; width: 10px; height: 10px;- border: 1.5px solid var(--line); border-top-color: var(--verdant);- border-radius: 50%; animation: spin .7s linear infinite;- }-- /* ---------- ask ---------- */- .ask-label {- font-size: var(--fs-label); letter-spacing: .12em; text-transform: uppercase;- color: var(--ink-faint); font-weight: 600; margin-bottom: var(--s2);- }- .ask-row { display: flex; gap: var(--s2); }- .ask-input {- flex: 1;- font-family: var(--serif); font-size: var(--fs-read);- padding: 11px 14px;- border: 1px solid var(--line); border-radius: var(--r-md);- background: var(--panel);- min-width: 0;- }- .ask-input::placeholder { color: var(--ink-faint); font-style: italic; }- .btn {- padding: 10px 18px; border-radius: var(--r-md); border: 1px solid var(--verdant-deep);- background: var(--verdant); color: #fff; font-weight: 600; font-size: 14px;- cursor: pointer;- transition: background var(--dur-fast), transform var(--dur-fast), box-shadow var(--dur-fast), border-color var(--dur-fast);- }- .btn:hover { background: var(--verdant-deep); transform: translateY(-1px); box-shadow: var(--shadow-1); }- .btn:active { transform: none; box-shadow: none; }- .btn:disabled { opacity: .55; cursor: default; transform: none; box-shadow: none; }- .btn.quiet { background: var(--panel); color: var(--ink); border-color: var(--line); font-weight: 500; }- .btn.quiet:hover { background: var(--panel); border-color: var(--ink-faint); }- .linklike { background: none; border: none; color: var(--verdant-deep); cursor: pointer; font-size: var(--fs-ui); padding: var(--s1) 0; }- .linklike:hover { text-decoration: underline; }+ /* ---------- page ---------- */+ .page { max-width: 760px; margin: 0 auto; padding: var(--s8) var(--s5) var(--s8); }+ section { margin-bottom: var(--s8); }- .local-note {- display: flex; gap: var(--s3); align-items: flex-start;- margin-top: var(--s3); padding: 9px 12px; border-radius: var(--r-md);- background: var(--amber-soft); border: 1px solid #e5d7ba;- font-size: 13px; color: #6d5426;+ /* ---------- hero ---------- */+ .hero h1 {+ font-family: var(--serif); font-weight: 600; letter-spacing: -0.01em;+ font-size: clamp(30px, 6vw, 46px); line-height: 1.12; margin-bottom: var(--s5);}- .local-note code { font-family: var(--mono); font-size: 12px; }- .note-close { background: none; border: none; cursor: pointer; color: #6d5426; font-size: 15px; line-height: 1.3; padding: 0; flex: none; }+ .hero-sub { color: var(--ink-soft); font-size: 18px; line-height: 1.6; max-width: 56ch; }+ .hero-sub b { color: var(--ink); font-weight: 600; }- /* ---------- answer ---------- */- .answer-card {- margin-top: var(--s4); background: var(--panel);- border: 1px solid var(--line); border-radius: var(--r-lg);- padding: var(--s4) var(--s5);- box-shadow: var(--shadow-1);- }- .answer-q {- font-family: var(--mono); font-size: var(--fs-meta); color: var(--ink-faint);- margin-bottom: var(--s3); word-break: break-word;- }- .answer-body { font-family: var(--serif); font-size: var(--fs-read); line-height: 1.65; }- .answer-body p + p { margin-top: var(--s3); }- .answer-body ul { margin: var(--s2) 0 var(--s2) 22px; }- .answer-body .cite {- font-family: var(--mono); font-size: 10.5px; font-style: normal;- background: var(--verdant-soft); color: var(--verdant-deep);- border-radius: var(--r-sm); padding: 1px 5px; margin: 0 1px;- cursor: pointer; vertical-align: super; line-height: 1;- border: none;- transition: background var(--dur-fast), color var(--dur-fast);- }- .answer-body .cite:hover { background: var(--verdant); color: #fff; }- .thinking { font-family: var(--serif); font-style: italic; color: var(--ink-soft); }-- #answer-skeleton .skel-bar {- height: 13px; border-radius: var(--r-sm); margin: var(--s3) 0;- background: linear-gradient(90deg, #e8ede7 25%, #f4f7f3 45%, #e8ede7 65%);- background-size: 200% 100%;- animation: shimmer 1.1s linear infinite;- }-- /* ---------- the receipt (signature element) ---------- */- .receipt {- display: inline-flex; align-items: center; gap: 7px;- font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-soft);- border: 1px solid var(--line); border-radius: var(--r-sm);- padding: 2px 8px 2px 6px; background: var(--panel);- white-space: nowrap;- }- .receipt .meter {- width: 34px; height: 5px; border-radius: 3px; background: #e4e9e3;- overflow: hidden; flex: none;- }- .receipt .meter i { display: block; height: 100%; background: var(--verdant); border-radius: 3px; }- .receipt.low .meter i { background: var(--amber); }- .receipt .pct { color: var(--ink); font-weight: 600; }-- #receipt-legend {- display: flex; gap: var(--s3); align-items: center; flex-wrap: wrap;- margin-top: var(--s3); padding: var(--s3) var(--s4);- background: var(--verdant-soft); border: 1px solid #cfe0d2; border-radius: var(--r-md);- font-size: var(--fs-ui); color: var(--ink-soft);- }- #receipt-legend .legend-text { flex: 1; min-width: 200px; }- #receipt-legend .linklike { font-family: var(--mono); font-size: var(--fs-meta); flex: none; }-- /* ---------- evidence tabs ---------- */- #evidence-tabs {- display: flex; gap: var(--s4);- border-bottom: 1px solid var(--line);- margin: var(--s5) 0 var(--s3);- }- .tab {- background: none; border: none; cursor: pointer;- font-size: var(--fs-label); letter-spacing: .12em; text-transform: uppercase; font-weight: 600;- color: var(--ink-faint);- padding: var(--s2) 2px; position: relative;- transition: color var(--dur-fast);- }- .tab .count { font-family: var(--mono); letter-spacing: 0; margin-left: 4px; color: inherit; font-weight: 500; }- .tab::after {- content: ""; position: absolute; left: 0; right: 0; bottom: -1px; height: 2px;- background: var(--verdant); transform: scaleX(0);- transition: transform var(--dur-fast) var(--ease-out);+ /* ---------- metric tiles ---------- */+ .metrics { display: grid; grid-template-columns: repeat(3, 1fr); gap: var(--s3); margin-top: var(--s6); }+ .metric {+ background: var(--panel); border: 1px solid var(--line); border-radius: var(--r-lg);+ padding: var(--s4) var(--s4) var(--s3); box-shadow: var(--shadow-1);}- .tab:hover { color: var(--ink); }- .tab.active { color: var(--ink); }- .tab.active::after { transform: scaleX(1); }+ .metric .n { font-family: var(--serif); font-weight: 600; font-size: 34px; color: var(--verdant-deep); line-height: 1; }+ .metric .l { font-family: var(--mono); font-size: 11.5px; color: var(--ink-soft); margin-top: var(--s2); text-transform: uppercase; letter-spacing: .06em; }- /* ---------- ledger sections ---------- */- .ledger-head {- display: flex; align-items: baseline; justify-content: space-between;- margin: 0 0 var(--s3);+ /* ---------- steps ---------- */+ h2.head {+ font-size: 12px; letter-spacing: .12em; text-transform: uppercase;+ color: var(--ink-faint); font-weight: 700; margin-bottom: var(--s5);border-bottom: 1px solid var(--line); padding-bottom: var(--s2);}- .ledger-head h2 {- font-size: var(--fs-label); letter-spacing: .12em; text-transform: uppercase;- color: var(--ink-faint); font-weight: 600;- }- .ledger-head .count { font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-faint); }-- .fact {- display: flex; flex-direction: column; gap: 5px;- padding: 11px 2px; border-bottom: 1px solid var(--line);- }- .fact-top { display: flex; align-items: baseline; gap: var(--s2); flex-wrap: wrap; }- .fact-name { font-family: var(--serif); font-size: 15.5px; font-weight: 700; cursor: pointer; }- .fact-name:hover { color: var(--verdant-deep); text-decoration: underline; }- .fact-type { font-family: var(--mono); font-size: 10.5px; color: var(--ink-faint); text-transform: lowercase; }- .fact-via { font-family: var(--mono); font-size: 10px; color: var(--verdant-deep); background: var(--verdant-soft); border-radius: 3px; padding: 1px 5px; }- .fact-via.expanded { color: var(--ink-soft); background: var(--paper); }- .fact-summary { font-family: var(--serif); font-size: 14px; color: var(--ink-soft); line-height: 1.5; }-- .edge-line { font-family: var(--mono); font-size: var(--fs-meta); color: var(--ink-soft); padding: 6px 2px; border-bottom: 1px solid var(--line); }- .edge-line b { color: var(--ink); font-weight: 600; }- .edge-line .rel { color: var(--verdant-deep); }-- .source {- padding: var(--s3) 2px; border-bottom: 1px solid var(--line);- }- .source.flash { background: var(--verdant-soft); transition: background 1.2s; }- .source-top { display: flex; align-items: baseline; gap: var(--s2); margin-bottom: 5px; }- .source-num { font-family: var(--mono); font-size: var(--fs-label); font-weight: 700; color: var(--verdant-deep); }- .source-title { font-size: 13px; font-weight: 600; }- .source-text {- font-family: var(--serif); font-size: 13.5px; color: var(--ink-soft); line-height: 1.55;- max-height: 88px; overflow: hidden; position: relative;- }- .source.open .source-text { max-height: none; }- .source-more { font-size: var(--fs-meta); color: var(--verdant-deep); background: none; border: none; cursor: pointer; padding: 2px 0; }-- /* ---------- library drawers ---------- */- details.grow { margin-bottom: var(--s3); border: 1px solid var(--line); border-radius: var(--r-lg); background: var(--panel); }- details.grow summary {- cursor: pointer; padding: var(--s3) var(--s4); font-weight: 600; font-size: var(--fs-ui);- list-style: none; display: flex; align-items: center; gap: var(--s2);- }- details.grow summary::-webkit-details-marker { display: none; }- details.grow summary::before { content: "+"; font-family: var(--mono); color: var(--verdant); font-weight: 700; }- details.grow[open] summary::before { content: "–"; }- details.grow .grow-body { padding: 2px var(--s4) var(--s4); }- .grow-hint { font-size: 12.5px; color: var(--ink-soft); margin-bottom: var(--s3); }- textarea.teach {- width: 100%; min-height: 74px; resize: vertical;- font-family: var(--serif); font-size: 15px; padding: 10px 12px;- border: 1px solid var(--line); border-radius: var(--r-md); background: var(--paper);- margin-bottom: var(--s2);- }- input.dir {- width: 100%; font-family: var(--mono); font-size: 13px; padding: 9px 12px;- border: 1px solid var(--line); border-radius: var(--r-md); background: var(--paper);- margin-bottom: var(--s2);- }- .grow-status { font-family: var(--mono); font-size: var(--fs-meta); margin-top: var(--s3); color: var(--ink-soft); white-space: pre-wrap; word-break: break-word; }- .grow-status:empty { margin-top: 0; }- .grow-status.ok { color: var(--verdant-deep); }- .grow-status.err { color: #a03b2e; }- .picker-row { display: flex; flex-wrap: wrap; gap: var(--s2); }- details.server-path { margin-top: var(--s3); border-top: 1px solid var(--line); padding-top: 10px; }- details.server-path summary { cursor: pointer; font-size: 12.5px; color: var(--ink-faint); list-style: none; }- details.server-path summary::-webkit-details-marker { display: none; }- details.server-path summary::before { content: "›"; margin-right: 6px; color: var(--verdant); }- details.server-path[open] summary::before { content: "⌄"; }- details.server-path .grow-hint code { font-family: var(--mono); font-size: 11.5px; }- details.server-path[open] { padding-bottom: var(--s1); }- details.server-path > *:not(summary) { margin-top: var(--s2); }- .watch-opt {- display: flex; align-items: center; gap: 7px; font-size: 12.5px; color: var(--ink-soft);- cursor: pointer; user-select: none;+ .steps { list-style: none; display: flex; flex-direction: column; gap: var(--s5); }+ .steps li { display: flex; gap: var(--s4); align-items: flex-start; }+ .steps .num {+ font-family: var(--mono); font-size: 12px; font-weight: 700;+ color: var(--verdant-deep); background: var(--verdant-soft); border-radius: 50%;+ width: 26px; height: 26px; flex: none;+ display: inline-flex; align-items: center; justify-content: center; margin-top: 2px;}- .watch-opt input { accent-color: var(--verdant); }+ .steps .body { font-size: 16px; color: var(--ink-soft); line-height: 1.55; }+ .steps .body b { color: var(--ink); font-weight: 600; }- .sync-pill {- display: flex; align-items: center; gap: 8px; margin-top: var(--s3);- padding: 8px 12px; border: 1px solid var(--verdant); border-radius: var(--r-sm);- background: var(--verdant-soft); font-size: 12.5px; color: var(--ink-soft);- }- .sync-pill .dot {- width: 8px; height: 8px; border-radius: 50%; background: var(--verdant); flex: none;- animation: sync-pulse 2s ease-in-out infinite;+ /* ---------- code block ---------- */+ pre {+ background: var(--ink); color: #e7ede8; border-radius: var(--r-lg);+ padding: var(--s5); overflow-x: auto; font-family: var(--mono); font-size: 13.5px; line-height: 1.7;}- .sync-pill #folder-sync-text { flex: 1; min-width: 0; }- .sync-pill .btn { padding: 3px 10px; font-size: 12px; }- @keyframes sync-pulse { 0%, 100% { opacity: 1; } 50% { opacity: .35; } }+ pre .c { color: #7f9686; } /* comment */+ .after { color: var(--ink-soft); font-size: 15px; margin-top: var(--s4); line-height: 1.6; }+ .after code { background: var(--verdant-soft); color: var(--verdant-deep); border-radius: var(--r-sm); padding: 1px 5px; font-size: 13px; }- .doc-row {- display: flex; gap: var(--s2); align-items: baseline; padding: 7px 4px;- border-bottom: 1px solid var(--line); font-size: 13px; border-radius: var(--r-sm);- transition: background var(--dur-fast);- }- .doc-row:hover { background: var(--verdant-soft); }- .doc-row .t { font-weight: 600; flex: 1; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }- .doc-row .d { font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-faint); flex: none; }+ /* ---------- prose ---------- */+ .prose { font-size: 16px; color: var(--ink-soft); line-height: 1.65; max-width: 62ch; }+ .prose + .prose { margin-top: var(--s3); }+ .prose b { color: var(--ink); font-weight: 600; }- /* ---------- graph panel chrome ---------- */- .atlas-legend {- position: absolute; top: 14px; left: 16px; z-index: 5;- font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-soft);- background: color-mix(in srgb, var(--panel) 88%, transparent);- border: 1px solid var(--line); border-radius: var(--r-md); padding: var(--s2) var(--s3);- display: flex; gap: 14px; align-items: center;+ footer {+ border-top: 1px solid var(--line); padding: var(--s5) 0 0;+ color: var(--ink-faint); font-size: 13px;}- .atlas-legend .swatch { display: inline-block; width: 9px; height: 9px; border-radius: 50%; margin-right: 5px; vertical-align: -1px; }- #node-card {- position: absolute; bottom: 16px; left: 16px; right: 16px; z-index: 5;- max-width: 440px;- background: var(--panel); border: 1px solid var(--line); border-radius: var(--r-lg);- padding: 14px 16px; box-shadow: var(--shadow-2);- visibility: hidden; opacity: 0; transform: translateY(6px) scale(.98);- transition: opacity var(--dur-fast) var(--ease-out), transform var(--dur-fast) var(--ease-out), visibility var(--dur-fast);- }- #node-card.show { visibility: visible; opacity: 1; transform: none; }- #node-card .close { position: absolute; top: 8px; right: 10px; background: none; border: none; cursor: pointer; color: var(--ink-faint); font-size: 15px; }- #node-card h3 { font-family: var(--serif); font-size: 17px; margin-bottom: 2px; }- #node-card .nc-type { font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-faint); margin-bottom: 7px; }- #node-card .nc-summary { font-family: var(--serif); font-size: 14px; color: var(--ink-soft); line-height: 1.5; margin-bottom: 9px; }- #node-card .nc-prov { font-family: var(--mono); font-size: var(--fs-label); color: var(--ink-faint); margin-top: var(--s2); }- .atlas-empty {- position: absolute; inset: 0; display: flex; align-items: center; justify-content: center;- z-index: 4; text-align: center; padding: 40px;- }- .atlas-empty .inner { max-width: 360px; }- .atlas-empty h3 { font-family: var(--serif); font-size: var(--fs-title); margin-bottom: var(--s2); }- .atlas-empty p { color: var(--ink-soft); font-size: 14px; }- .hidden { display: none !important; }-- @media (max-width: 900px) {- .frame { grid-template-columns: 1fr; grid-template-rows: auto 420px; height: auto; }- .desk { border-right: none; border-bottom: 1px solid var(--line); }- .atlas { height: 420px; }- }- @media (prefers-reduced-motion: reduce) {- * { transition: none !important; animation: none !important; }- .rise { opacity: 1 !important; transform: none !important; }- #answer-skeleton .skel-bar { background: #e8ede7; }- #node-card { visibility: hidden; opacity: 1; transform: none; }- #node-card.show { visibility: visible; }+ @media (max-width: 560px) {+ .metrics { grid-template-columns: 1fr; }+ .page { padding: var(--s6) var(--s4) var(--s7); }}-- /* ---------- access-token gate ---------- */- .token-overlay {- position: fixed; inset: 0; z-index: 1000;- display: none; align-items: center; justify-content: center;- background: rgba(28, 36, 34, .55); padding: 20px;- }- .token-overlay.open { display: flex; }- .token-card {- width: 100%; max-width: 440px;- background: var(--panel); border: 1px solid var(--line); border-radius: var(--r-lg);- padding: 24px 24px 20px; box-shadow: 0 12px 40px rgba(28, 36, 34, .25);- }- .token-card h2 {- font-family: var(--serif); font-size: var(--fs-title); font-weight: 600; margin-bottom: var(--s2);- }- .token-card p { font-size: 14px; color: var(--ink-soft); margin-bottom: 14px; }- .token-card p code {- font-family: var(--mono); font-size: var(--fs-meta);- background: var(--paper); border: 1px solid var(--line);- border-radius: var(--r-sm); padding: 1px 5px; color: var(--ink);- }- .token-input {- width: 100%; padding: 11px 13px; font-family: var(--mono); font-size: 13px;- border: 1px solid var(--line); border-radius: var(--r-md); background: var(--panel);- }- .token-input:focus { border-color: var(--verdant); }- .token-hint { font-size: 12.5px; color: var(--ink-faint); margin: 10px 0 0; }- .token-hint code { font-family: var(--mono); font-size: 11.5px; }- .token-err { color: var(--amber); font-size: 13px; margin-top: 10px; min-height: 0; }- .token-actions { display: flex; justify-content: flex-end; gap: 10px; margin-top: 18px; }</style></head><body>-<div class="token-overlay" id="token-overlay" role="dialog" aria-modal="true" aria-labelledby="token-title">- <div class="token-card">- <h2 id="token-title">This graph is protected</h2>- <p>Paste the <b>access token</b> or the <b>share link</b> you were given.</p>- <input class="token-input" id="token-input" type="text"- placeholder="token, or http://…/#token=…" autocomplete="off" spellcheck="false" />- <div class="token-err" id="token-err"></div>- <p class="token-hint">- Where to find it: open <code>./data/share-link.txt</code> next to the server,- or run <code>docker compose logs | grep token</code>.- </p>- <div class="token-actions">- <button class="btn quiet" id="token-cancel" type="button">Cancel</button>- <button class="btn" id="token-submit" type="button">Unlock</button>- </div>- </div>-</div>-<header class="masthead"><span class="wordmark">Context<span class="dot">·</span>Graph</span>- <span class="tagline">a living ledger of what your team knows</span>- <div class="masthead-right">- <span class="stat"><b id="stat-nodes">–</b> facts</span>- <span class="stat-sep">·</span>- <span class="stat"><b id="stat-edges">–</b> connections</span>- <span class="stat-sep">·</span>- <span class="stat"><b id="stat-docs">–</b> documents</span>- <span class="provider-badge" id="provider-badge">…</span>- </div>+ <span class="tagline">read the graph before you work</span></header>-<div class="frame">- <main class="desk">-- <!-- Onboarding hero — shown while the graph is empty -->- <section id="onboarding" class="hero rise" style="--i:1">- <h1>Start your ledger</h1>- <p class="hero-sub">Add documents and Context Graph turns them into a map of facts —- every answer cited, every fact carrying a receipt that shows how well-attested it is.- <em>Agents that read the graph before a task explore 31% less.</em></p>- <ol class="steps">- <li><span class="num">1</span><span><b>Add documents</b> — PDFs, markdown, or plain text.</span></li>- <li><span class="num">2</span><span><b>Ask a question</b> — get a written answer with numbered citations.</span></li>- <li><span class="num">3</span><span><b>Check the receipts</b> — every fact shows its confidence and sources.</span></li>- </ol>- <div id="onboarding-controls-slot">- <div id="ingest-controls">- <button class="dropzone" id="dropzone" type="button">- <span class="dz-main">Drop files here, or click to choose</span>- <span class="dz-sub">.pdf · .md · .txt — files stay on your machine</span>- </button>- <input type="file" id="files-input" multiple accept=".pdf,.md,.markdown,.txt" hidden />- <input type="file" id="folder-input" webkitdirectory multiple hidden />- <div class="picker-row">- <button class="btn quiet" id="choose-folder-btn" type="button" title="The folder keeps syncing while this tab is open">Connect a folder…</button>- </div>- <div id="folder-sync" class="sync-pill hidden">- <span class="dot"></span>- <span id="folder-sync-text"></span>- <button class="btn quiet" id="folder-sync-stop" type="button">Stop</button>- </div>- <details class="server-path">- <summary>Ingest a path on the server</summary>- <p class="grow-hint">Reads a file or folder from the server's own disk — in Docker, that's the mounted <code>/docs</code>.</p>- <input class="dir" id="ingest-input" placeholder="/path/to/docs or /path/to/file.md" />- <label class="watch-opt"><input type="checkbox" id="watch-check" /> Keep watching — future changes are ingested automatically</label>- <button class="btn quiet" id="ingest-btn" type="button">Ingest server path</button>- </details>- <div id="ingest-progress" class="hidden">- <div class="progress-track"><i id="progress-fill"></i></div>- <div id="progress-rows"></div>- </div>- <div class="grow-status" id="ingest-status"></div>- </div>- </div>- </section>-- <!-- Ask — the hero once documents exist -->- <section id="ask-section" class="rise" style="--i:1">- <div class="ask-label">Ask your documents</div>- <form class="ask-row" id="ask-form">- <input class="ask-input" id="ask-input" placeholder="How do we deploy to staging?" autocomplete="off" />- <button class="btn" id="ask-btn" type="submit">Ask</button>- </form>- <div class="local-note hidden" id="local-note">- <span>Local mode — search and the map work now. For written answers and adding new- documents, run a free local model (<code>ollama pull llama3.2</code>) or set- <code>OPENROUTER_API_KEY</code>.</span>- <button class="note-close" id="local-note-close" type="button" aria-label="Dismiss">×</button>- </div>- </section>-- <section id="answer-wrap" class="hidden">- <div class="answer-card">- <div class="answer-q" id="answer-q"></div>- <div id="answer-skeleton" class="hidden">- <span class="thinking">Consulting the graph…</span>- <div class="skel-bar" style="width:92%"></div>- <div class="skel-bar" style="width:78%"></div>- <div class="skel-bar" style="width:60%"></div>- </div>- <div class="answer-body" id="answer-body"></div>- </div>-- <div id="receipt-legend" class="hidden">- <span id="legend-sample"></span>- <span class="legend-text">The receipt: how confident the graph is (87%), how many- times the fact was observed (×4), and in how many sources (2). Seen often = trusted more.</span>- <button class="linklike" id="legend-dismiss" type="button">got it ×</button>- </div>-- <div id="evidence" class="hidden">- <nav id="evidence-tabs">- <button type="button" class="tab active" data-tab="sources">Sources <span class="count" id="sources-count"></span></button>- <button type="button" class="tab" data-tab="facts">Facts <span class="count" id="facts-count"></span></button>- <button type="button" class="tab" data-tab="edges" id="tab-btn-edges">Connections <span class="count" id="edges-count"></span></button>- </nav>- <div class="tab-panel" id="tab-sources"><div id="sources"></div></div>- <div class="tab-panel hidden" id="tab-facts"><div id="facts"></div></div>- <div class="tab-panel hidden" id="tab-edges"><div id="edges"></div></div>- </div>- </section>-- <!-- Library — documents + ways to add knowledge -->- <section id="library" class="rise" style="--i:2">- <div class="ledger-head"><h2>Library</h2><span class="count" id="docs-count"></span></div>-- <details class="grow" id="ingest-panel">- <summary>Add documents</summary>- <div class="grow-body" id="ingest-panel-body">- <p class="grow-hint">Every file becomes facts and connections — watch the map grow on the right.</p>- </div>- </details>-- <details class="grow" id="teach-panel">- <summary>Add a fact</summary>- <div class="grow-body">- <p class="grow-hint">Know something the documents don't say? Write it in one sentence —- “Postgres failovers page the on-call via #incidents, not email.” The graph will- remember it, and who said so.</p>- <textarea class="teach" id="teach-input" placeholder="What did you learn?"></textarea>- <button class="btn quiet" id="teach-btn" type="button">Save fact</button>- <div class="grow-status" id="teach-status"></div>- </div>- </details>-- <div id="docs"></div>- <button class="linklike hidden" id="docs-toggle" type="button">Show all</button>- </section>- </main>-- <section class="atlas">- <div class="atlas-legend">- <span><span class="swatch" style="background:#2e6b4e"></span>used in this answer</span>- <span><span class="swatch" style="background:#b8c2ba"></span>everything else</span>- </div>- <div id="graph"></div>- <div class="atlas-empty hidden" id="atlas-empty">- <div class="inner">- <h3>Your knowledge map draws itself here.</h3>- <p>As documents are added, each fact becomes a node; each relationship, a line.</p>- </div>- </div>- <div id="node-card">- <button class="close" id="node-card-close" aria-label="Close">×</button>- <h3 id="nc-name"></h3>- <div class="nc-type" id="nc-type"></div>- <div class="nc-summary" id="nc-summary"></div>- <span class="receipt" id="nc-receipt"></span>- <div class="nc-prov" id="nc-prov"></div>+<main class="page">++ <section class="hero rise" style="--i:0">+ <h1>31% fewer tool calls.<br>23% less cost.<br>17% lower latency.</h1>+ <p class="hero-sub">That's what your agent gets from reading the repo's context graph before it+ starts working. The graph is a folder of plain markdown files in the repo — one per system,+ API, or concept, linked to each other. It stays in sync with the code through git, so+ <b>anyone who clones the repo has it, no setup.</b></p>+ <div class="metrics">+ <div class="metric"><div class="n">−31%</div><div class="l">tool calls</div></div>+ <div class="metric"><div class="n">−23%</div><div class="l">cost</div></div>+ <div class="metric"><div class="n">−17%</div><div class="l">latency</div></div></div></section>-</div>--<script src="https://unpkg.com/[email protected]/standalone/umd/vis-network.min.js" integrity="sha384-yxKDWWf0wwdUj/gPeuL11czrnKFQROnLgY8ll7En9NYoXibgg3C6NK/UDHNtUgWJ" crossorigin="anonymous"></script>-<script>-"use strict";-const $ = (id) => document.getElementById(id);-const REDUCED = matchMedia("(prefers-reduced-motion: reduce)");--/* ---------- access token (shared/team deployments) ---------- */-// A share link carries #token=… in the URL fragment — fragments are never-// sent to the server, so the token stays out of proxy/access logs. Accept-// either a bare token or a whole pasted link and pull the token out of it.-function extractToken(raw) {- const s = (raw || "").trim();- if (!s) return "";- // Full share link: token lives in the #token=… fragment.- const hash = s.includes("#") ? s.slice(s.indexOf("#") + 1) : "";- const fromHash = new URLSearchParams(hash).get("token");- return (fromHash || s).trim();-}--(function pickUpToken() {- const frag = new URLSearchParams(location.hash.replace(/^#/, ""));- const t = frag.get("token");- if (t) {- localStorage.setItem("cg-token", t.trim());- history.replaceState(null, "", location.pathname + location.search);- }-})();--// In-page token modal — a Promise that resolves with the entered token (already-// normalized) or null if the user cancels. Replaces the browser's window.prompt.-function askForToken(message) {- const overlay = $("token-overlay");- const input = $("token-input");- const err = $("token-err");- err.textContent = message || "";- input.value = "";- overlay.classList.add("open");- input.focus();- return new Promise((resolve) => {- const close = (value) => {- overlay.classList.remove("open");- $("token-submit").onclick = null;- $("token-cancel").onclick = null;- input.onkeydown = null;- overlay.onmousedown = null;- resolve(value);- };- const submit = () => {- const token = extractToken(input.value);- if (!token) { err.textContent = "Enter a token or share link."; return; }- close(token);- };- $("token-submit").onclick = submit;- $("token-cancel").onclick = () => close(null);- input.onkeydown = (e) => {- if (e.key === "Enter") submit();- else if (e.key === "Escape") close(null);- };- // Click on the backdrop (not the card) cancels.- overlay.onmousedown = (e) => { if (e.target === overlay) close(null); };- });-}--async function api(path, opts = {}) {- const headers = { ...(opts.headers || {}) };- const token = localStorage.getItem("cg-token");- if (token) headers["authorization"] = "Bearer " + token;- const r = await fetch(path, { ...opts, headers });- if (r.status === 401) {- const wrong = Boolean(token); // had a token that the server rejected- const t = await askForToken(wrong ? "That token was rejected. Try again." : "");- if (t) {- localStorage.setItem("cg-token", t);- return api(path, opts);- }- throw new Error("access token required");- }- return r;-}-const esc = (s) => String(s ?? "").replace(/[&<>"']/g, (c) => ({ "&": "&", "<": "<", ">": ">", '"': """, "'": "'" }[c]));--const MODEL_HINT = "Adding facts needs a model: run Ollama locally (ollama pull llama3.2) or set OPENROUTER_API_KEY.";--const COLOR_BASE = { background: "#b8c2ba", border: "#93a098" };-const COLOR_HIT = { background: "#2e6b4e", border: "#1f4f38" };-const COLOR_EXPANDED = { background: "#7fa98f", border: "#2e6b4e" };-const COLOR_FRESH = { background: "#3f8f66", border: "#1f4f38" };--let network = null;-let nodesDs = null;-let edgesDs = null;-let prevObs = new Map(); // node id -> observations, for "what just got smarter" diffing-let highlighted = new Set();-let ingestBusy = false;--/* ---------- app state: empty (onboarding) vs ready (ask) ---------- */-let appState = "";-function setAppState(stats) {- const forcedReady = new URLSearchParams(location.search).get("q");- const ready = (stats && stats.documents > 0) || Boolean(forcedReady);- const next = ready ? "is-ready" : "is-empty";- if (next === appState) return;- appState = next;- document.body.classList.remove("is-ready", "is-empty");- document.body.classList.add(next);- // The ingest controls exist once; they live in the hero while the graph is- // empty and move into the Library drawer once it isn't.- const controls = $("ingest-controls");- if (ready) {- $("ingest-panel-body").appendChild(controls);- if (ingestBusy) {- // Mid-ingest flip: keep the progress visible in its new home.- $("ingest-panel").open = true;- } else {- $("ask-input").focus();- }- } else {- $("onboarding-controls-slot").appendChild(controls);- }-}--/* ---------- motion helpers ---------- */-// Stagger-reveal freshly rendered children (they're new DOM nodes each render,-// so the animation replays). Skipped entirely under reduced motion.-function stagger(el, cap = 10) {- if (REDUCED.matches) return;- Array.from(el.children).forEach((c, i) => {- c.style.setProperty("--i", String(Math.min(i, cap)));- c.classList.add("rise");- });-}--/* ---------- receipts ---------- */-function receiptHtml(confidence, observations, sourceCount) {- const pct = Math.round(confidence * 100);- const low = pct < 55;- const src = sourceCount != null ? ` · ${sourceCount} source${sourceCount === 1 ? "" : "s"}` : "";- const across = sourceCount != null ? ` across ${sourceCount} source${sourceCount === 1 ? "" : "s"}` : "";- return `<span class="receipt${low ? " low" : ""}" title="confidence ${pct}% — observed ${observations} time${observations === 1 ? "" : "s"}${esc(across)}; repeated observations raise confidence">` +- `<span class="meter"><i style="width:${pct}%"></i></span>` +- `<span class="pct">${pct}%</span><span>×${observations}${esc(src)}</span></span>`;-}--function maybeShowReceiptLegend() {- if (localStorage.getItem("cg-receipt-legend")) return;- $("receipt-legend").classList.remove("hidden");-}-$("legend-dismiss").addEventListener("click", () => {- localStorage.setItem("cg-receipt-legend", "1");- $("receipt-legend").classList.add("hidden");-});--/* ---------- stats + provider ---------- */-async function refreshStats() {- const s = await (await api("/api/stats")).json();- $("stat-nodes").textContent = s.nodes;- $("stat-edges").textContent = s.edges;- $("stat-docs").textContent = s.documents;- const badge = $("provider-badge");- if (s.provider.canSynthesize) {- badge.textContent = "answers: " + s.provider.model;- badge.classList.add("cloud");- } else {- badge.textContent = "local mode";- if (!localStorage.getItem("cg-local-note-off")) $("local-note").classList.remove("hidden");- }- setAppState(s);- return s;-}-$("local-note-close").addEventListener("click", () => {- localStorage.setItem("cg-local-note-off", "1");- $("local-note").classList.add("hidden");-});-$("provider-badge").addEventListener("click", () => {- // In local mode the badge re-opens the dismissed explainer.- if ($("provider-badge").classList.contains("cloud")) return;- localStorage.removeItem("cg-local-note-off");- $("local-note").classList.remove("hidden");-});--/* ---------- graph ---------- */-function nodeVisual(n) {- return {- id: n.id,- label: n.name,- title: `${n.name} — ${n.type}\n${n.summary}\n×${n.observations} observations · ${Math.round(n.confidence * 100)}% confidence`,- size: 9 + Math.min(n.observations, 10) * 1.8,- color: highlighted.has(n.id) ? COLOR_HIT : COLOR_BASE,- _data: n,- };-}--async function refreshGraph({ diff = false } = {}) {- const g = await (await api("/api/graph")).json();- $("atlas-empty").classList.toggle("hidden", g.nodes.length > 0);-- const freshIds = [];- if (diff) {- for (const n of g.nodes) {- const before = prevObs.get(n.id);- if (before === undefined || n.observations > before) freshIds.push(n.id);- }- }- prevObs = new Map(g.nodes.map((n) => [n.id, n.observations]));-- const nodes = g.nodes.map(nodeVisual);- const edges = g.edges.map((e, i) => ({- id: "e" + i,- from: e.sourceId,- to: e.targetId,- title: `${e.relation}${e.description ? " — " + e.description : ""}\n×${e.observations} · ${Math.round(e.confidence * 100)}%`,- width: 1 + Math.min(e.observations - 1, 4) * 0.6,- }));-- if (!network) {- nodesDs = new vis.DataSet(nodes);- edgesDs = new vis.DataSet(edges);- network = new vis.Network($("graph"), { nodes: nodesDs, edges: edgesDs }, {- nodes: {- shape: "dot",- font: { face: "system-ui", size: 12, color: "#1c2422" },- borderWidth: 1.5,- },- edges: {- color: { color: "#c9d1c9", highlight: "#2e6b4e", hover: "#8b968f" },- arrows: { to: { enabled: true, scaleFactor: 0.5 } },- smooth: { type: "continuous" },- },- interaction: { hover: true, tooltipDelay: 120 },- physics: {- solver: "forceAtlas2Based",- forceAtlas2Based: { gravitationalConstant: -42, springLength: 110, damping: 0.5 },- stabilization: { iterations: 180 },- },- });- network.on("click", (params) => {- if (params.nodes.length) showNodeCard(params.nodes[0]);- else hideNodeCard();- });- } else {- // Update in place so the layout doesn't reshuffle under the user.- const existing = new Set(nodesDs.getIds());- for (const n of nodes) existing.has(n.id) ? nodesDs.update(n) : nodesDs.add(n);- edgesDs.clear();- edgesDs.add(edges);- }-- if (freshIds.length) pulseFresh(freshIds);- return g;-}--function pulseFresh(ids) {- nodesDs.update(ids.map((id) => ({ id, color: COLOR_FRESH, borderWidth: 3 })));- if (network) network.fit({ nodes: ids, animation: REDUCED.matches ? false : { duration: 600 }, maxZoomLevel: 1.2 });- setTimeout(() => {- nodesDs.update(ids.map((id) => ({- id,- color: highlighted.has(id) ? COLOR_HIT : COLOR_BASE,- borderWidth: 1.5,- })));- }, 2600);-}--function highlightRetrieval(bundle) {- highlighted = new Set(bundle.nodes.map((n) => n.id));- const semantic = new Set(bundle.nodes.filter((n) => n.via === "semantic").map((n) => n.id));- nodesDs.update(nodesDs.getIds().map((id) => ({- id,- color: semantic.has(id) ? COLOR_HIT : highlighted.has(id) ? COLOR_EXPANDED : COLOR_BASE,- opacity: highlighted.has(id) ? 1 : 0.35,- })));- if (highlighted.size && network) {- network.fit({ nodes: [...highlighted], animation: REDUCED.matches ? false : { duration: 700 }, maxZoomLevel: 1.1 });- }-}--function focusNode(id) {- if (!network) return;- network.focus(id, { scale: 1.3, animation: REDUCED.matches ? false : { duration: 500 } });- network.selectNodes([id]);- showNodeCard(id);-}--function showNodeCard(id) {- const v = nodesDs.get(id);- if (!v || !v._data) return;- const n = v._data;- $("nc-name").textContent = n.name;- $("nc-type").textContent = n.type;- $("nc-summary").textContent = n.summary;- $("nc-receipt").outerHTML = receiptHtml(n.confidence, n.observations).replace('class="receipt', 'id="nc-receipt" class="receipt');- $("nc-prov").textContent = "";- $("node-card").classList.add("show");-}-function hideNodeCard() { $("node-card").classList.remove("show"); }-$("node-card-close").addEventListener("click", hideNodeCard);--/* ---------- evidence tabs ---------- */-function activateTab(name) {- document.querySelectorAll("#evidence-tabs .tab").forEach((b) =>- b.classList.toggle("active", b.dataset.tab === name));- ["sources", "facts", "edges"].forEach((t) =>- $("tab-" + t).classList.toggle("hidden", t !== name));-}-document.querySelectorAll("#evidence-tabs .tab").forEach((b) =>- b.addEventListener("click", () => activateTab(b.dataset.tab)));-/* ---------- ask ---------- */-function renderAnswerText(text, sourceCount) {- // Minimal, safe rendering: escape first, then paragraphs, bullets, bold, [n] citations.- const lines = esc(text).split(/\n/);- let html = "", inList = false;- const flushList = () => { if (inList) { html += "</ul>"; inList = false; } };- let para = [];- const flushPara = () => {- if (para.length) { html += "<p>" + para.join("<br>") + "</p>"; para = []; }- };- for (const raw of lines) {- const line = raw.trim();- if (!line) { flushPara(); flushList(); continue; }- if (/^[-*•]\s+/.test(line)) {- flushPara();- if (!inList) { html += "<ul>"; inList = true; }- html += "<li>" + line.replace(/^[-*•]\s+/, "") + "</li>";- } else {- flushList();- para.push(line);- }- }- flushPara(); flushList();- html = html.replace(/\*\*([^*]+)\*\*/g, "<strong>$1</strong>");- html = html.replace(/\[(\d+)\]/g, (m, d) => {- const i = Number(d);- return i >= 1 && i <= sourceCount ? `<button class="cite" data-src="${i}">${i}</button>` : m;- });- return html;-}--function renderBundle(bundle) {- const facts = $("facts");- facts.innerHTML = bundle.nodes.map((n) => `- <div class="fact">- <div class="fact-top">- <span class="fact-name" data-node="${esc(n.id)}">${esc(n.name)}</span>- <span class="fact-type">${esc(n.type)}</span>- <span class="fact-via${n.via === "expanded" ? " expanded" : ""}">${n.via === "semantic" ? "matched" : "connected"}</span>- ${receiptHtml(n.confidence, n.observations, n.provenance.length)}- </div>- <div class="fact-summary">${esc(n.summary)}</div>- </div>`).join("");- $("facts-count").textContent = bundle.nodes.length;- facts.querySelectorAll(".fact-name").forEach((el) =>- el.addEventListener("click", () => focusNode(el.dataset.node)));-- const byId = new Map(bundle.nodes.map((n) => [n.id, n.name]));- const edgeRows = bundle.edges- .filter((e) => byId.has(e.sourceId) && byId.has(e.targetId))- .slice(0, 10)- .map((e) => `<div class="edge-line"><b>${esc(byId.get(e.sourceId))}</b> <span class="rel">—${esc(e.relation)}→</span> <b>${esc(byId.get(e.targetId))}</b></div>`);- $("edges").innerHTML = edgeRows.join("");- $("edges-count").textContent = edgeRows.length;- $("tab-btn-edges").classList.toggle("hidden", !edgeRows.length);-- $("sources").innerHTML = bundle.chunks.map((c, i) => `- <div class="source" id="source-${i + 1}">- <div class="source-top">- <span class="source-num">[${i + 1}]</span>- <span class="source-title">${esc(c.documentTitle)}</span>- </div>- <div class="source-text">${esc(c.text)}</div>- <button class="source-more" type="button">show more</button>- </div>`).join("");- $("sources-count").textContent = bundle.chunks.length;- $("sources").querySelectorAll(".source-more").forEach((btn) =>- btn.addEventListener("click", () => {- const card = btn.closest(".source");- card.classList.toggle("open");- btn.textContent = card.classList.contains("open") ? "show less" : "show more";- }));-- activateTab("sources");- // Cascade the evidence in after the prose (see stagger()).- stagger($("sources"));- stagger($("facts"));- stagger($("edges"));-}--$("ask-form").addEventListener("submit", async (ev) => {- ev.preventDefault();- const q = $("ask-input").value.trim();- if (!q) return;- const btn = $("ask-btn");- btn.disabled = true;- $("answer-wrap").classList.remove("hidden");- $("answer-q").textContent = "Q · " + q;- $("answer-body").innerHTML = "";- $("answer-skeleton").classList.remove("hidden");- $("evidence").classList.add("hidden");- $("facts").innerHTML = ""; $("edges").innerHTML = ""; $("sources").innerHTML = "";- try {- const r = await api("/api/ask", {- method: "POST",- headers: { "content-type": "application/json" },- body: JSON.stringify({ question: q }),- });- const data = await r.json();- if (data.error) throw new Error(data.error);- const { answer, bundle } = data;- $("answer-skeleton").classList.add("hidden");- const hasEvidence = bundle.nodes.length > 0 || bundle.chunks.length > 0;- if (!hasEvidence) {- $("answer-body").innerHTML = '<p class="thinking">Nothing in your documents answers this yet. <button class="linklike" id="no-results-add" type="button">Add the documents</button> that would — then ask again.</p>';- const add = $("no-results-add");- if (add) add.addEventListener("click", () => {- $("ingest-panel").open = true;- $("ingest-panel").scrollIntoView({ behavior: REDUCED.matches ? "auto" : "smooth", block: "center" });- });- } else if (answer) {- $("answer-body").innerHTML = renderAnswerText(answer, bundle.chunks.length);- $("answer-body").querySelectorAll(".cite").forEach((el) =>- el.addEventListener("click", () => {- activateTab("sources");- const card = $("source-" + el.dataset.src);- if (!card) return;- card.scrollIntoView({ behavior: REDUCED.matches ? "auto" : "smooth", block: "center" });- card.classList.add("flash");- setTimeout(() => card.classList.remove("flash"), 1400);- }));- } else {- $("answer-body").innerHTML = '<p class="thinking">Local mode — here is everything your documents hold on this, with receipts. For a written answer, run Ollama or set OPENROUTER_API_KEY.</p>';- }- stagger($("answer-body"), 4);- $("evidence").classList.toggle("hidden", !hasEvidence);- renderBundle(bundle);- highlightRetrieval(bundle);- if (hasEvidence) maybeShowReceiptLegend();- } catch (e) {- $("answer-skeleton").classList.add("hidden");- $("answer-body").innerHTML = '<p class="thinking">Something went wrong: ' + esc(e.message) + "</p>";- } finally {- btn.disabled = false;- }-});--/* ---------- teach ---------- */-$("teach-btn").addEventListener("click", async () => {- const learning = $("teach-input").value.trim();- const status = $("teach-status");- if (!learning) return;- $("teach-btn").disabled = true;- status.className = "grow-status";- status.textContent = "Extracting entities and folding them in…";- try {- const r = await api("/api/contribute", {- method: "POST",- headers: { "content-type": "application/json" },- body: JSON.stringify({ learning }),- });- const data = await r.json();- if (data.error) throw new Error(data.error);- status.className = "grow-status ok";- status.textContent =- `✓ saved — ${data.nodesCreated} new fact${data.nodesCreated === 1 ? "" : "s"}, ` +- `${data.nodesUpdated} reinforced, ${data.edgesCreated + data.edgesUpdated} connections touched`;- $("teach-input").value = "";- await refreshStats();- await refreshGraph({ diff: true });- await refreshDocs();- } catch (e) {- status.className = "grow-status err";- status.textContent = "Could not save: " + e.message + "\n" + MODEL_HINT;- } finally {- $("teach-btn").disabled = false;- }-});--/* ---------- ingest progress (shared by server-path and upload) ---------- */-let progressItems = [];-function progressShow() {- progressItems = [];- $("progress-fill").style.width = "0%";- $("progress-rows").innerHTML = "";- $("ingest-progress").classList.remove("hidden");-}-function renderProgress(frac) {- if (frac != null) $("progress-fill").style.width = Math.round(frac * 100) + "%";- $("progress-rows").innerHTML = progressItems.slice(-4).map((it) => `- <div class="progress-row ${it.state}">- <span class="icon">${it.state === "busy" ? '<span class="spinner"></span>' : it.state === "ok" ? "✓" : "✗"}</span>- <span>${esc(it.name)}${it.msg ? " — " + esc(it.msg) : ""}</span>- </div>`).join("");-}-function progressFile(index, total, name) {- for (const it of progressItems) if (it.state === "busy") it.state = "ok";- progressItems.push({ name, state: "busy" });- renderProgress(total ? index / total : 0);-}-function progressFail(msg) {- const last = progressItems[progressItems.length - 1];- if (last) { last.state = "err"; last.msg = msg; }- renderProgress(null);-}-function progressDone() {- for (const it of progressItems) if (it.state === "busy") it.state = "ok";- renderProgress(1);-}--function setIngestControlsDisabled(disabled) {- ingestBusy = disabled;- [$("dropzone"), $("choose-folder-btn"), $("ingest-btn")].forEach((b) => { b.disabled = disabled; });-}--/* ---------- ingest a path on the server ---------- */-$("ingest-btn").addEventListener("click", async () => {- const dir = $("ingest-input").value.trim();- const status = $("ingest-status");- if (!dir) return;- setIngestControlsDisabled(true);- status.className = "grow-status";- status.textContent = "Scanning " + dir + " …";- progressShow();- try {- const r = await api("/api/ingest-dir", {- method: "POST",- headers: { "content-type": "application/json" },- body: JSON.stringify({ dir, watch: $("watch-check").checked }),- });- if (!r.ok) {- const err = await r.json().catch(() => ({}));- throw new Error(err.error || r.statusText);- }- const reader = r.body.getReader();- const decoder = new TextDecoder();- let buf = "";- while (true) {- const { done, value } = await reader.read();- if (done) break;- buf += decoder.decode(value, { stream: true });- let nl;- while ((nl = buf.indexOf("\n")) >= 0) {- const line = buf.slice(0, nl); buf = buf.slice(nl + 1);- if (!line.trim()) continue;- const ev = JSON.parse(line);- if (ev.type === "progress") {- status.textContent = "";- progressFile(ev.index, ev.total, ev.file.split("/").pop());- // Grow the graph live between files.- refreshGraph({ diff: true }).catch(() => {});- refreshStats().catch(() => {});- } else if (ev.type === "done") {- progressDone();- const fresh = ev.results.filter((x) => !x.skipped);- const skipped = ev.results.length - fresh.length;- const nodes = fresh.reduce((a, x) => a + x.nodesCreated + x.nodesUpdated, 0);- status.className = "grow-status ok";- status.textContent =- `✓ ${fresh.length} document${fresh.length === 1 ? "" : "s"} added` +- (skipped ? ` (${skipped} already known, skipped)` : "") +- `, ${nodes} facts written or reinforced` +- (ev.watching ? " — now watching this folder for changes" : "");- } else if (ev.type === "error") {- throw new Error(ev.error);- }- }- }- await refreshStats();- await refreshGraph({ diff: true });- await refreshDocs();- } catch (e) {- progressFail(e.message);- status.className = "grow-status err";- status.textContent = "Could not add documents: " + e.message + "\n" + MODEL_HINT;- } finally {- setIngestControlsDisabled(false);- }-});--/* ---------- browser upload (dropzone / pickers → server ingest) ---------- */-const UPLOAD_ACCEPT = /\.(pdf|md|markdown|txt)$/i;--// Base64-encode file bytes in chunks (fromCharCode.apply chokes on huge arrays).-// The relative path (from a folder pick or live sync) keys the document on the-// server, so same-named files in different subfolders stay distinct.-async function fileToPayload(file) {- const relPath = file._relPath || file.webkitRelativePath || undefined;- if (/\.pdf$/i.test(file.name)) {- const bytes = new Uint8Array(await file.arrayBuffer());- let binary = "";- for (let i = 0; i < bytes.length; i += 0x8000) {- binary += String.fromCharCode.apply(null, bytes.subarray(i, i + 0x8000));- }- return { name: file.name, relPath, dataBase64: btoa(binary) };- }- return { name: file.name, relPath, text: await file.text() };-}--$("dropzone").addEventListener("click", () => $("files-input").click());-$("dropzone").addEventListener("dragover", (e) => {- e.preventDefault();- if (!ingestBusy) $("dropzone").classList.add("drag");-});-$("dropzone").addEventListener("dragleave", () => $("dropzone").classList.remove("drag"));-$("dropzone").addEventListener("drop", (e) => {- e.preventDefault();- $("dropzone").classList.remove("drag");- if (ingestBusy) return;- uploadFiles(Array.from(e.dataTransfer.files || []), "files");-});-$("choose-folder-btn").addEventListener("click", async () => {- // Chromium: keep a handle to the picked folder and live-sync it while the- // tab is open. Elsewhere: fall back to the classic one-shot folder upload.- if (!window.showDirectoryPicker) return $("folder-input").click();- try {- const handle = await showDirectoryPicker({ mode: "read" });- await startFolderSync(handle);- } catch (e) {- if (e.name !== "AbortError") {- $("ingest-status").className = "grow-status err";- $("ingest-status").textContent = "Could not open the folder: " + e.message;- }- }-});-$("files-input").addEventListener("change", (e) => uploadPicked(e, "files"));-$("folder-input").addEventListener("change", (e) => uploadPicked(e, "folder"));--function uploadPicked(e, kind) {- const all = Array.from(e.target.files || []);- e.target.value = ""; // let the user re-pick the same selection later- uploadFiles(all, kind);-}--async function uploadFiles(all, kind) {- const files = all.filter((f) => UPLOAD_ACCEPT.test(f.name));- const status = $("ingest-status");- if (!files.length) {- status.className = "grow-status err";- status.textContent = all.length- ? `No .pdf, .md, or .txt files in that ${kind === "folder" ? "folder" : "selection"}.`- : `No ${kind === "folder" ? "folder" : "files"} selected.`;- return;- }- setIngestControlsDisabled(true);- status.className = "grow-status";- status.textContent = "";- progressShow();- let ok = 0, skipped = 0, failed = 0, nodes = 0;- try {- for (let i = 0; i < files.length; i++) {- const f = files[i];- progressFile(i, files.length, f.name);- try {- const r = await api("/api/ingest-upload", {- method: "POST",- headers: { "content-type": "application/json" },- body: JSON.stringify(await fileToPayload(f)),- });- if (!r.ok) {- const er = await r.json().catch(() => ({}));- throw new Error(er.error || r.statusText);- }- const data = await r.json();- if (data.result && !data.result.skipped) {- ok++;- nodes += (data.result.nodesCreated || 0) + (data.result.nodesUpdated || 0);- } else {- skipped++;- }- // Grow the graph live between files, like the server-folder ingest.- refreshGraph({ diff: true }).catch(() => {});- refreshStats().catch(() => {});- } catch (err) {- failed++;- progressFail(err.message);- }- }- progressDone();- status.className = failed ? "grow-status err" : "grow-status ok";- status.textContent =- `✓ ${ok} document${ok === 1 ? "" : "s"} added` +- (skipped ? ` (${skipped} already known, skipped)` : "") +- (failed ? `, ${failed} failed` : "") +- `, ${nodes} facts written or reinforced` +- (failed ? "\n" + MODEL_HINT : "");- await refreshStats();- await refreshGraph({ diff: true });- await refreshDocs();- } finally {- setIngestControlsDisabled(false);- }-}--/* ---------- live folder sync (File System Access API, Chromium) ----------- The picked folder's handle stays with the page: every few seconds it is- re-scanned and files whose size/mtime moved are re-uploaded, so the graph- keeps evolving while the tab is open. Deletions are ignored on purpose —- the graph is append-only. */-const SYNC_INTERVAL_MS = 5000;-const SYNC_SKIP = new Set(["node_modules", "dist", "build", "out", "target", "vendor", "coverage", "__pycache__", "venv"]);-const SYNC_MAX_BYTES = 1_000_000;-let folderSync = null; // { handle, name, state: Map<rel, sig>, timer, busy }--async function collectFolderFiles(handle, name) {- const files = [];- const state = new Map();- async function walk(dir, prefix) {- for await (const entry of dir.values()) {- if (entry.name.startsWith(".") || SYNC_SKIP.has(entry.name)) continue;- if (entry.kind === "directory") {- await walk(entry, prefix + entry.name + "/");- } else if (UPLOAD_ACCEPT.test(entry.name)) {- const f = await entry.getFile();- if (f.size > SYNC_MAX_BYTES) continue;- const rel = prefix + entry.name;- state.set(rel, f.size + ":" + f.lastModified);- f._relPath = name + "/" + rel;- f._rel = rel;- files.push(f);- }- }- }- await walk(handle, "");- return { files, state };-}+ <section class="rise" style="--i:1">+ <h2 class="head">How it works</h2>+ <ol class="steps">+ <li>+ <span class="num">1</span>+ <span class="body"><b>Point it at your repo.</b> It builds <code>.context/</code> — one+ markdown file per system, API, or concept, linked together. The files are the nodes, the+ links are the edges. That's the whole graph.</span>+ </li>+ <li>+ <span class="num">2</span>+ <span class="body"><b>Commit it.</b> The graph is just files in git, so git is the sync.+ Anyone who clones the repo has it, and CI keeps it fresh — when a change moves something,+ you see it in the graph diff right next to the code, in the same PR.</span>+ </li>+ <li>+ <span class="num">3</span>+ <span class="body"><b>Your agent reads it before doing work</b> — so it starts knowing the+ codebase instead of rediscovering it from scratch every time.</span>+ </li>+ </ol>+ </section>-function renderSyncPill(msg) {- $("folder-sync-text").textContent = msg;- $("folder-sync").classList.remove("hidden");-}+ <section class="rise" style="--i:2">+ <h2 class="head">Setup</h2>+<pre><span class="c"># in your repo</span>+context-graph init <span class="c"># builds .context/ from your code</span>+git add .context && git commit-async function startFolderSync(handle) {- stopFolderSync();- const session = { handle, name: handle.name, state: new Map(), timer: null, busy: false };- folderSync = session;- renderSyncPill(`Scanning “${handle.name}”…`);- const { files, state } = await collectFolderFiles(handle, handle.name);- session.state = state;- await uploadFiles(files, "folder"); // the usual progress UI for the first sweep- if (folderSync !== session) return; // stopped or replaced mid-sweep- session.timer = setInterval(syncTick, SYNC_INTERVAL_MS);- renderSyncPill(`Watching “${handle.name}” — ${state.size} files. Edits sync while this tab is open.`);-}+<span class="c"># keep it honest — fails CI if the graph is stale</span>+context-graph check</pre>+ <p class="after">That's it. Anyone who clones the repo gets <code>.context/</code> for free.+ Point your agent at it and it reads the graph before working.</p>+ </section>-function stopFolderSync() {- if (!folderSync) return;- clearInterval(folderSync.timer);- folderSync = null;- $("folder-sync").classList.add("hidden");-}-$("folder-sync-stop").addEventListener("click", stopFolderSync);+ <section class="rise" style="--i:3">+ <h2 class="head">The proof</h2>+ <p class="prose">Same model, same tools, same tasks. The only difference is whether the agent+ got the graph first. On a real 344-file repo it cut tool calls 31%, cost 23%, and latency 17%.</p>+ <p class="prose">The graph isn't magic — it's just that reading a few linked files beats+ rediscovering the codebase from scratch every single time.</p>+ </section>-async function syncTick() {- const session = folderSync;- if (!session || session.busy || ingestBusy) return;- session.busy = true;- try {- const { files, state } = await collectFolderFiles(session.handle, session.name);- const changed = files.filter((f) => session.state.get(f._rel) !== state.get(f._rel));- session.state = state; // forgets deleted entries too — harmless, append-only- let synced = 0;- for (const f of changed) {- if (folderSync !== session) return; // stopped mid-batch- const r = await api("/api/ingest-upload", {- method: "POST",- headers: { "content-type": "application/json" },- body: JSON.stringify(await fileToPayload(f)),- });- if (r.ok) {- const data = await r.json();- if (data.result && !data.result.skipped) synced++;- }- }- if (synced && folderSync === session) {- refreshStats().catch(() => {});- refreshGraph({ diff: true }).catch(() => {});- refreshDocs().catch(() => {});- const at = new Date().toLocaleTimeString();- renderSyncPill(`Watching “${session.name}” — synced ${synced} change${synced === 1 ? "" : "s"} at ${at}.`);- }- } catch (e) {- // Permission revoked or the folder went away — stop cleanly, not noisily.- if (folderSync === session) {- clearInterval(session.timer);- folderSync = null;- renderSyncPill(`Sync stopped: ${e.message}`);- }- return;- }- session.busy = false;-}+ <section class="rise" style="--i:4">+ <h2 class="head">Runs locally</h2>+ <p class="prose">Everything runs on your machine. No accounts, no cloud calls, no keys. Building+ the graph uses a local model (Ollama); if you'd rather use a cloud model for better summaries,+ set a key and it'll use it. Local stays the default.</p>+ </section>-/* ---------- documents ---------- */-const DOCS_PREVIEW = 10;-let docsExpanded = false;-let docsCache = [];-function renderDocs() {- const shown = docsExpanded ? docsCache : docsCache.slice(0, DOCS_PREVIEW);- $("docs").innerHTML = shown.map((d) => `- <div class="doc-row">- <span class="t" title="${esc(d.source)}">${esc(d.title)}</span>- <span class="d">${esc((d.createdAt || "").slice(0, 10))}</span>- </div>`).join("");- const toggle = $("docs-toggle");- toggle.classList.toggle("hidden", docsCache.length <= DOCS_PREVIEW);- toggle.textContent = docsExpanded ? "Show fewer" : `Show all ${docsCache.length}`;-}-$("docs-toggle").addEventListener("click", () => {- docsExpanded = !docsExpanded;- renderDocs();-});-async function refreshDocs() {- docsCache = await (await api("/api/documents")).json();- $("docs-count").textContent = docsCache.length;- renderDocs();- if (docsCache.length && !$("ask-input").value) {- const t = docsCache[0].title.length > 42 ? docsCache[0].title.slice(0, 42) + "…" : docsCache[0].title;- $("ask-input").placeholder = `What does “${t}” cover?`;- }-}+ <footer>Context·Graph — a context graph that lives in your repo.</footer>-/* ---------- boot ---------- */-(async function boot() {- $("legend-sample").innerHTML = receiptHtml(0.87, 4, 2);- try {- await refreshStats();- if (typeof vis === "undefined") {- $("graph").innerHTML = '<p style="padding:40px;color:#5c6862;font-family:var(--serif)">The graph view needs internet access once (vis-network CDN). Everything else works offline.</p>';- } else {- await refreshGraph();- }- await refreshDocs();- } catch (e) {- console.error(e);- // Stats unavailable (e.g. token prompt cancelled) — default to the ask- // layout so the page still works once access is sorted out.- if (!appState) setAppState({ documents: 1 });- }- // Shareable question links: /?q=how+do+we+deploy auto-asks on load.- const q = new URLSearchParams(location.search).get("q");- if (q) {- $("ask-input").value = q;- $("ask-form").requestSubmit();- }-})();+</main>-/* ---------- live updates ----------- When the server runs with an auto-watcher (CONTEXT_GRAPH_WATCH=1), the graph- grows on its own as watched folders change. A cheap stats poll notices and- refreshes the heavier graph/docs views only when the counts actually moved. */-let liveCounts = null;-setInterval(async () => {- try {- const s = await refreshStats();- const counts = `${s.documents}|${s.nodes}|${s.edges}|${s.chunks}`;- if (liveCounts !== null && counts !== liveCounts) {- refreshGraph({ diff: true }).catch(() => {});- refreshDocs().catch(() => {});- }- liveCounts = counts;- } catch {- /* offline or token prompt pending — try again next tick */- }-}, 10_000);-</script></body></html>