Skip to content

Self-improving AI Employees for everyone

docs(cli_agent): describe the pool-only Claude path, UUID resume, and the CLI pin

trohitga913f04f7631
diff --git a/CLAUDE.md b/CLAUDE.md
index e41331a8..4f27fcaa 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -64,8 +64,8 @@ This is a React Flow-based workflow automation platform implementing n8n-inspire
| **[Status Broadcaster](./docs-internal/status_broadcaster.md)** | WebSocket-first communication: StatusBroadcaster singleton, live count via `len(MESSAGE_HANDLERS) + len(get_ws_handlers())`, broadcast message types, Android two-state model |
| **[RLM Service](./docs-internal/rlm_service.md)** | Recursive Language Model agent with REPL-based execution (llm_query, rlm_query, FINAL) |
| **[Claude Code Agent](./docs-internal/claude_code_agent.md)** | Hub for the Claude Code agent — routes to architecture, interactive mode, CLI framework, canonical-patterns RFC, and the 5 CLI snapshot references. |
-| **[CLI Agent Framework](./docs-internal/cli_agent_framework.md)** | Multi-provider CLI runtime (Claude Code / Codex / Gemini): `AICliService.run_batch`, per-task worktree isolation, FastMCP bridge, **memory bridge** (`--continue` for first run + intra-process stream-json multi-turn + `--resume <UUID>` for crash recovery, all on a stable `cwd=repo_root`; `node_parameters_updated` broadcast on every successful turn). **Plugin-folder layout** (post-cutover): all claude-specific code lives in [`server/nodes/agent/claude_code_agent/`](./server/nodes/agent/claude_code_agent/) — `_provider.py`, `_pool.py`, `_skills.py`, `_oauth.py`, `_handlers.py` — and self-registers via three `factory.py` registries (`register_provider` / `register_session_pool` / `register_skill_materialiser`) plus `register_ws_handlers`. The generic framework at `services/cli_agent/` imports nothing from `nodes/`. |
-| **[Claude Code Interactive Mode](./docs-internal/claude_code_interactive_mode.md)** | The interactive-mode cutover: OpenCompany no longer uses `claude -p` headless. `ClaudeSessionPool` (keyed by `simpleMemory.node_id`) spawns `claude` as a plain subprocess with stdio pipes — **no PTY** — and drives it over the VSCode-extension protocol: `--output-format stream-json --input-format stream-json --verbose --ide`. User prompts written as JSON to `proc.stdin` (`{"type":"user","message":{"role":"user","content":"..."}}\n`); `system/init` / `assistant` / `result` / `system/compact_boundary` events stream back on `proc.stdout` (parsed by a background `stdout_reader_task` — the on-disk JSONL is persistence-only, not the runtime contract, because `result` events are stdout-only in stream-json mode). Cross-platform via plain pipes (the earlier PTY pattern was broken on Windows because pywinpty/ConPTY's emulated stdin never reached claude's Ink TUI keystroke handler). Stays in interactive billing — entrypoint `claude-vscode`, NOT `sdk-cli`. Multi-turn within one warm subprocess preserves the session UUID (verified end-to-end); cross-batch continuity via `--continue` (first run) or `--resume <UUID>` (crash recovery). Four typed CloudEvents (`claude.session.{spawned,cleared,terminated,usage}`) fire from the pool. `/compact` events forward to `CompactionService`. The non-pooled `AICliSession` path (one-shot prompt-in-argv runs) still uses PTY (out of scope for this refactor; works on POSIX). |
+| **[CLI Agent Framework](./docs-internal/cli_agent_framework.md)** | Multi-provider CLI runtime (Claude Code / Codex / Gemini): `AICliService.run_batch`, per-task worktree isolation, FastMCP bridge, **memory bridge** (host-minted `--session-id` on cold spawn + intra-process stream-json multi-turn + `--resume <last_session_id>` for memory-bound runs and crash recovery — never `--continue`, which the CLI resolves only against interactive sessions — all on a stable `cwd=repo_root`; `node_parameters_updated` broadcast on every successful turn). **Plugin-folder layout** (post-cutover): all claude-specific code lives in [`server/nodes/agent/claude_code_agent/`](./server/nodes/agent/claude_code_agent/) — `_provider.py`, `_pool.py`, `_skills.py`, `_oauth.py`, `_handlers.py` — and self-registers via three `factory.py` registries (`register_provider` / `register_session_pool` / `register_skill_materialiser`) plus `register_ws_handlers`. The generic framework at `services/cli_agent/` imports nothing from `nodes/`. |
+| **[Claude Code Interactive Mode](./docs-internal/claude_code_interactive_mode.md)** | The interactive-mode cutover: OpenCompany no longer uses `claude -p` headless. `ClaudeSessionPool` (keyed by `simpleMemory.node_id`) spawns `claude` as a plain subprocess with stdio pipes — **no PTY** — and drives it over the VSCode-extension protocol: `--output-format stream-json --input-format stream-json --verbose --ide`. User prompts written as JSON to `proc.stdin` (`{"type":"user","message":{"role":"user","content":"..."}}\n`); `system/init` / `assistant` / `result` / `system/compact_boundary` events stream back on `proc.stdout` (parsed by a background `stdout_reader_task` — the on-disk JSONL is persistence-only, not the runtime contract, because `result` events are stdout-only in stream-json mode). Cross-platform via plain pipes (the earlier PTY pattern was broken on Windows because pywinpty/ConPTY's emulated stdin never reached claude's Ink TUI keystroke handler). Stays in interactive billing — entrypoint `claude-vscode`, NOT `sdk-cli`. Multi-turn within one warm subprocess preserves the session UUID (verified end-to-end); cross-batch continuity via `--continue` (first run) or `--resume <UUID>` (crash recovery). Four typed CloudEvents (`claude.session.{spawned,cleared,terminated,usage}`) fire from the pool. `/compact` events forward to `CompactionService`. Every Claude run routes through the pool (unbound runs get an ephemeral session per task, terminated after the batch); `AICliSession` (PTY + JSONL watcher) is no longer used for Claude because a PTY stdin makes the CLI reject stream-json input and that path never delivered the prompt (GitHub #133 / #134). A child exiting without a `result` wakes `send_turn` on stdout EOF with the exit code instead of burning the turn timeout. The CLI version is pinned via `package_version` in `config/ai_cli_providers.json`. |
| **[CLI Agent Canonical Patterns RFC](./docs-internal/cli_agent_canonical_patterns_rfc.md)** | Audit of OpenCompany's `services/cli_agent` against the official Claude Code spec — six invariants (skills-as-files, MCP-only transport, `list_changed`, deferral, visible-tool filtering, native-session-continuity-via-stable-cwd) with current implementation status. |
| **[Claude Code CLI Reference (snapshot)](./docs-internal/claude_code_cli_reference.md)** | Verbatim snapshot of [code.claude.com/docs/en/cli-reference](https://code.claude.com/docs/en/cli-reference) — every CLI subcommand + flag + system-prompt-flag matrix + the subset we emit from `nodes/agent/claude_code_agent/_provider.py`. Fetched 2026-05-11. |
| **[Claude Code Env Vars (snapshot)](./docs-internal/claude_code_env_vars_reference.md)** | Categorised [code.claude.com/docs/en/env-vars](https://code.claude.com/docs/en/env-vars) snapshot — auth, Bedrock/Vertex, model, bash, MCP, telemetry, session/debug, paths. Documents `CLAUDE_CONFIG_DIR`, `MAX_MCP_OUTPUT_TOKENS`, `ENABLE_TOOL_SEARCH` which we touch directly. |
diff --git a/docs-internal/claude_code_agent.md b/docs-internal/claude_code_agent.md
index 1c3d6af1..d52736ab 100644
--- a/docs-internal/claude_code_agent.md
+++ b/docs-internal/claude_code_agent.md
@@ -6,7 +6,7 @@ stream-json protocol — not the `claude -p` headless path. A warm `ClaudeSessio
(keyed by the RFC-0002 conversation key when a Context node is connected, or by the
legacy `simpleMemory.node_id` on immutable V1 graphs — `services/cli_agent/service.py:354-358`)
preserves the session across turns;
-memory continuity is claude-native via `--continue` / `--resume <UUID>` on a stable
+memory continuity is claude-native via a host-minted `--session-id` on the first run and `--resume <last_session_id>` afterwards on a stable
`cwd=repo_root`. Connected tools and skills reach claude through an MCP bridge; the spawn
runs under `--permission-mode dontAsk` with a strict `--allowedTools` allowlist and stays
in interactive billing (entrypoint `claude-vscode`, not `sdk-cli`).
@@ -19,7 +19,7 @@ This page is a **router**. The detail lives in the documents below — start her
|---|---|
| How does the whole integration work end-to-end? (architecture, memory & context compaction, sub-agent creation, data structures) | [claude_code_agent_architecture.md](./claude_code_agent_architecture.md) |
| How does interactive mode work? (stdio pipes vs PTY, stream-json events, `system/init` / `assistant` / `result`, session continuity, the four `claude.session.*` CloudEvents) | [claude_code_interactive_mode.md](./claude_code_interactive_mode.md) |
-| The generic multi-provider runtime (`AICliService.run_batch`, worktree isolation, FastMCP bridge, the memory bridge `--continue`/`--resume`, plugin-folder layout) | [cli_agent_framework.md](./cli_agent_framework.md) |
+| The generic multi-provider runtime (`AICliService.run_batch`, FastMCP bridge, the memory bridge `--session-id`/`--resume`, plugin-folder layout) | [cli_agent_framework.md](./cli_agent_framework.md) |
| Does our implementation match the official Claude Code spec? (six invariants + status) | [cli_agent_canonical_patterns_rfc.md](./cli_agent_canonical_patterns_rfc.md) |
| Which CLI subcommands/flags exist, and which subset we emit | [claude_code_cli_reference.md](./claude_code_cli_reference.md) (snapshot) |
| Environment variables (`CLAUDE_CONFIG_DIR`, `MAX_MCP_OUTPUT_TOKENS`, `ENABLE_TOOL_SEARCH`, Bedrock/Vertex, telemetry) | [claude_code_env_vars_reference.md](./claude_code_env_vars_reference.md) (snapshot) |
@@ -39,7 +39,7 @@ All claude-specific code is self-contained here; the generic framework at
| File | Responsibility |
|---|---|
| `__init__.py` | `ClaudeCodeAgentNode(ActionNode)` (line 238 — it does not go through `SpecializedAgentBase` / the native loop) + `Params`/`Output`; self-registers via `factory.py` (`register_provider` / `register_session_pool` / `register_skill_materialiser`) + `register_ws_handlers`. |
-| `_provider.py` | Builds the spawn argv (`--output-format stream-json --input-format stream-json --verbose --ide`, `--permission-mode dontAsk`, `--allowedTools`, `--continue`/`--resume`). |
+| `_provider.py` | Builds the spawn argv (`--output-format stream-json --input-format stream-json --verbose --ide`, `--permission-mode dontAsk`, `--allowedTools`, `--resume`/`--session-id`). |
| `_pool.py` | `ClaudeSessionPool` — warm subprocess keyed by `simpleMemory.node_id`; stdout reader, session-UUID capture, crash-recovery respawn. |
| `_skills.py` | `materialise_skills` — writes connected SKILL.md trees under `<workspace>/.claude/skills/`, diff-based on warm reuse. |
| `_oauth.py` | Isolated `CLAUDE_CONFIG_DIR` (`<DATA_DIR>/claude/`), browser-OAuth bridge, binary install. |
@@ -47,5 +47,5 @@ All claude-specific code is self-contained here; the generic framework at
## See also
-- Memory bridge specifics (how `simpleMemory` feeds `--continue`, why `last_session_id` is display-only): [cli_agent_framework.md → Memory bridge](./cli_agent_framework.md).
+- Memory bridge specifics (how `simpleMemory.last_session_id` feeds `--resume`, why `--continue` is never used): [cli_agent_framework.md → Memory bridge](./cli_agent_framework.md).
- Generic CLI agent runtime shared with Codex / Gemini providers: `server/services/cli_agent/`.
diff --git a/docs-internal/claude_code_cli_reference.md b/docs-internal/claude_code_cli_reference.md
index 2e64a318..ea27ac5a 100644
--- a/docs-internal/claude_code_cli_reference.md
+++ b/docs-internal/claude_code_cli_reference.md
@@ -194,17 +194,21 @@ Quick cross-reference for the OpenCompany `claude_code_agent` integration:
| `--mcp-config <json>` | when MCP wired | inline JSON with `{mcpServers: {opencompany: {type: http, url, headers, alwaysLoad: true}}}` |
| `--strict-mcp-config` | when MCP wired | blocks user-level `~/.claude.json` MCP entries |
| `--append-system-prompt <text>` | when `system_prompt` is set | append the task prompt to Claude Code's built-in system prompt |
-| `--continue` | first spawn for a memory-bound run | let Claude select the latest conversation for the stable working directory |
-| `--resume <UUID>` | pooled-process crash or reap recovery | resume the session UUID captured from Claude's stream events |
+| `--session-id <UUID>` | cold spawn with no continuity flag | host-minted `uuid4` so the session UUID is known before the first event (must be a valid UUID; rejected if already in use) |
+| `--resume <UUID>` | memory-bound runs (`last_session_id` from the memory node) and pooled-process crash or reap recovery | resume by UUID; works from any directory since 2.1.223 |
+| `--continue` | never set by the node | the reference says it skips sessions created non-interactively (`-p` / stream-json), i.e. every session the pool creates; kept on `ClaudeTaskSpec` for explicit callers only |
| `--effort <level>` | optional per-task override | reasoning effort |
| `--add-dir <path>*` | optional per-task override | grant extra working dirs |
| `--disallowedTools <csv>` | optional per-task override | remove tools from context |
| `--agent <name>` | optional per-task override | use a specific subagent |
-The interactive subprocess path intentionally does not emit `--print`,
-`--include-partial-messages`, `--include-hook-events`, or `--session-id`.
-It also retains `max_turns`, `max_budget_usd`, and `fallback_model` on the
-task specification only for compatibility; their corresponding flags are
-print-mode-only and are not emitted by this path.
+The subprocess path intentionally does not emit `--print` (the official
+Agent SDKs omit it too; the help text's "only works with --print" note
+on the format flags predates that), `--include-partial-messages`, or
+`--include-hook-events`. It retains `max_turns`, `max_budget_usd`, and
+`fallback_model` on the task specification only for compatibility; their
+corresponding flags are print-mode-only and are not emitted by this path.
+The CLI version the argv is verified against is pinned in
+`server/config/ai_cli_providers.json` (`package_version`).
See [`server/nodes/agent/claude_code_agent/_provider.py`](../server/nodes/agent/claude_code_agent/_provider.py) for the argv-construction code.
diff --git a/docs-internal/claude_code_interactive_mode.md b/docs-internal/claude_code_interactive_mode.md
index 0539b566..848534d1 100644
--- a/docs-internal/claude_code_interactive_mode.md
+++ b/docs-internal/claude_code_interactive_mode.md
@@ -8,8 +8,8 @@
> --input-format stream-json --verbose --ide`. Multi-turn happens by
> writing newline-delimited JSON to `proc.stdin` of the long-lived
> subprocess. Events arrive on `proc.stdout`. The on-disk JSONL is no
-> longer the runtime contract — claude writes it for `--continue` /
-> `--resume` persistence but the `result` event is stdout-only in
+> longer the runtime contract — claude writes it for `--resume`
+> persistence but the `result` event is stdout-only in
> stream-json mode.
## Why we cut over from `-p`
@@ -66,8 +66,8 @@ server/nodes/agent/claude_code_agent/
server/services/cli_agent/ (generic framework — shared by all CLI plugins)
├── factory.py # register_provider / register_session_pool / register_skill_materialiser registries
-├── service.py # AICliService.run_batch — dispatcher; routes to pool when memory-bound (registry lookup)
-├── session.py # AICliSession — one-shot non-pooled path (still PTY on POSIX; out of scope)
+├── service.py # AICliService.run_batch — dispatcher; every claude task goes to the pool (registry lookup)
+├── session.py # AICliSession — generic PTY + JSONL path; NOT used for claude (GitHub #133 / #134)
├── mcp_server.py # BatchContext + FastMCP bridge (shared by every CLI provider)
├── workflow_tools.py / lockfile.py / jsonl_watcher.py / config.py / protocol.py / types.py / _cli_auth.py
└── _handlers.py # codex WS handlers only (will move once codex_agent migrates)
@@ -86,7 +86,7 @@ claude
--ide
[--mcp-config <json> --strict-mcp-config]
--model <name>
- [--resume <UUID> | --continue]
+ [--resume <UUID> | --continue | --session-id <UUID>]
--allowedTools <csv>
--permission-mode dontAsk
[--append-system-prompt <text>]
@@ -102,9 +102,16 @@ Notable absences (intentional):
|---|---|
| `-p` / `--print` | Drops us into the SDK billing bucket; the whole point of the cutover. |
| Positional prompt (`-- "<text>"`) | Prompt arrives via stdin in stream-json input mode; an argv positional would double-send the first turn. `interactive_argv`'s `include_prompt` parameter is kept for back-compat but ignored. |
-| `--session-id <UUID>` | Rejected by claude in non-headless mode. We discover the UUID from the first event that carries `session_id`. |
+| `--continue` | The CLI reference says it skips sessions created non-interactively (`-p` / stream-json input), which is exactly what the pool spawns, so it never found the prior conversation. Memory-bound runs pass `--resume <last_session_id>` instead (the UUID persisted on the memory node by `_persist_memory`). |
| `--include-partial-messages`, `--include-hook-events`, `--max-turns`, `--max-budget-usd`, `--fallback-model` | All `-p`-only. `ClaudeTaskSpec` keeps the fields for back-compat; they're silently dropped here. External cost / turn caps are a Phase 2 follow-up. |
+Emitted for session identity:
+
+| Flag | When |
+|---|---|
+| `--session-id <UUID>` | Cold spawn with no continuity flag. `ClaudeSessionPool._spawn` mints a `uuid4` so `current_session_uuid` is known before the first stream event; crash-recovery respawns then always have a UUID to `--resume`. The CLI requires a valid UUID and rejects one already in use. |
+| `--resume <UUID>` | Crash-recovery respawn of a pooled session, and every memory-bound run that has a persisted `last_session_id`. Works from any directory on the pinned CLI (2.1.223+). |
+
`--permission-mode dontAsk` is the documented mode for *"only
pre-approved tools, no prompts"* per
[`code.claude.com/docs/en/permission-modes`](https://code.claude.com/docs/en/permission-modes).
@@ -180,21 +187,24 @@ The pool's continuity model is two-layered:
- **Intra-process (warm pool).** Within a live subprocess, every turn
writes to the same `proc.stdin` and claude itself keeps the conversation
- in memory. No `--continue` / `--resume` needed — same session UUID
- flows across turns. Verified: turn 1 `13ac5819-...`, turn 2 same UUID.
-- **Cross-process (cold spawn / crash recovery).** Cold spawns of a
- memory-bound run emit `--continue` (claude auto-finds the latest
- conversation under the cwd via `project_key`). If `acquire` finds the
- pooled subprocess has died, it captures the dead session's UUID
- before respawning and splices `--resume <UUID>` into the spec so the
- new subprocess continues writing to the same on-disk JSONL.
-
-`ClaudeTaskSpec` carries `continue_session: bool` and
-`resume_session_id: Optional[str]`. `resume_session_id` wins if both
-are set. `simpleMemory.last_session_id` is display-only metadata —
-`_persist_memory` writes it for the UI but the agent doesn't read it
-back (continuity rides through `--continue` + the warm pool, not a
-ferried UUID).
+ in memory. No `--resume` needed — same session UUID flows across turns.
+ Verified: turn 1 `13ac5819-...`, turn 2 same UUID.
+- **Cross-process (cold spawn / crash recovery).** A cold spawn with no
+ continuity flag mints a `uuid4` and emits `--session-id <UUID>`, so
+ the UUID is known before the first event. Memory-bound runs emit
+ `--resume <last_session_id>` (the UUID `_persist_memory` stored on the
+ memory node after the previous successful run). If `acquire` finds the
+ pooled subprocess has died, it splices `--resume <current_session_uuid>`
+ into the spec so the new subprocess continues the same on-disk JSONL.
+
+`ClaudeTaskSpec` carries `session_id`, `resume_session_id` and
+`continue_session`; precedence in argv is resume, then continue, then
+session-id. `--continue` is never set by the node: the CLI reference says
+it skips sessions created non-interactively (`-p` / stream-json input),
+which is every session the pool creates, so it silently started a fresh
+conversation each batch. A stale `last_session_id` (claude reports
+`No conversation found with session ID`) is cleared by `_persist_memory`
+so the next run spawns fresh.
## MCP bearer token lifecycle
@@ -341,10 +351,10 @@ are copied wholesale via `shutil.copytree` so `scripts/` and
the UI) reconstruct the frontmatter (`name`, `description`,
`allowed-tools`, `metadata`) + the markdown body. Failure modes
are non-fatal: a skill that fails to load or write logs at WARN
-and is skipped — the spawn continues. Same helper is called from
-`AICliSession._pre_spawn` (non-pool path) and
-`ClaudeSessionPool._spawn` (pool path), so behaviour is uniform
-across the two transports.
+and is skipped — the spawn continues. The helper is called from
+`ClaudeSessionPool._spawn` (cold spawn) and `_prepare_warm_reuse`
+(diff on warm reuse); `AICliSession._pre_spawn` still calls it for the
+generic path other providers use.
**Migration note.** Pre-cutover code wrote SKILL.md trees into
`<repo_root>/.claude/skills/` so they accumulated in the user's
@@ -370,12 +380,11 @@ splices the per-workflow workspace
(`~/.opencompany/workspaces/<workflow_slug>/`, injected into `ctx.raw` by
`workflow.py:_get_workspace_dir`) into each task's `add_dir` list
right after `task_list` is built — `interactive_argv` already emits
-`--add-dir <path>` per entry. Runs BEFORE the pool/non-pool branch
-split so both warm-subprocess and one-shot paths get it. Without
-this, the workspace is invisible to claude: memory-bound runs spawn
-with `cwd=repo_root` (stable for `--continue`'s `project_key`
-resolution) and non-memory runs with `cwd=worktree`, neither of which
-sees files dropped by upstream nodes (`fileDownloader`,
+`--add-dir <path>` per entry. Runs BEFORE the pool branch so every
+pooled session gets it. Without this, the workspace is invisible to
+claude: sessions spawn with `cwd=repo_root` (stable for `--resume`'s
+`project_key` resolution), which does not see files dropped by
+upstream nodes (`fileDownloader`,
`documentParser`, code executors, etc.). Mirrors the ai_agent
pattern ([`services/ai.py:1186`](../server/services/ai.py) in
`execute_agent`, `:1924` in `execute_chat_agent` —
@@ -409,6 +418,14 @@ on-disk session JSONL via `JsonlWatcher` / `JsonlDirWatcher`. That path
was abandoned for the pool: `pywinpty`'s ConPTY emulation did not
deliver keystrokes to claude's Ink TUI on Windows (empirically
confirmed across four test variants). Stdio pipes + stream-json work
-cross-platform and match the Anthropic-blessed pattern. The non-pooled
-`AICliSession` still uses a PTY on POSIX for its one-shot
-`AICliService.run_batch` path — out of scope for this refactor.
+cross-platform and match the pattern the official Agent SDKs use. The
+non-pooled `AICliSession` was kept as the plain-run path for a while,
+but it never wrote the prompt to the child and a PTY stdin makes the
+CLI reject `--input-format stream-json` (GitHub #133 / #134), so
+`AICliService.run_batch` now routes every Claude task through the pool:
+bound runs get a warm keyed session, unbound runs get an ephemeral
+session per task (`<node_id>:<turn_execution_id>:<index>`) terminated
+after the batch. A child that exits without a `result` wakes
+`send_turn` on stdout EOF with its exit code instead of burning the
+turn timeout. The CLI version is pinned via `package_version` in
+`config/ai_cli_providers.json` (currently 2.1.258).
diff --git a/docs-internal/cli_agent_canonical_patterns_rfc.md b/docs-internal/cli_agent_canonical_patterns_rfc.md
index aa4051dd..3a1154ca 100644
--- a/docs-internal/cli_agent_canonical_patterns_rfc.md
+++ b/docs-internal/cli_agent_canonical_patterns_rfc.md
@@ -288,8 +288,9 @@ stream-json`, `--verbose`, `--ide`, `--model`, `--permission-mode`,
`--allowedTools <csv>`.
**Conditional:** `--mcp-config <json>` + `--strict-mcp-config` (when
-`mcp_endpoint_url` and `mcp_bearer_token` set), `--continue`/`--resume`
-(memory-bound runs, mutually exclusive), `--append-system-prompt`,
+`mcp_endpoint_url` and `mcp_bearer_token` set), `--resume` /
+`--session-id` (memory-bound runs / cold spawn, mutually exclusive with
+each other and with `--continue`), `--append-system-prompt`,
`--effort`, `--add-dir`, `--disallowedTools`, `--agent`.
**Never emitted** (`_provider.py:316-320`): `--max-turns`,
@@ -416,11 +417,13 @@ Either works.
### 4.8 Memory bridge — `simpleMemory` → `claude_code_agent` (DONE in `ecbe69b`)
**Status:** native claude session continuity works end-to-end. The
-mechanism was later simplified from the UUID5 / `--session-id`
-round-trip described in this section to `--continue` (first cold spawn
-for a memory-wired run) + intra-process stream-json multi-turn against a
-warm subprocess + `--resume <UUID>` for crash recovery — all on a stable
-`cwd=repo_root`. The entry point is
+mechanism is a host-minted `--session-id <uuid4>` on the first cold
+spawn + intra-process stream-json multi-turn against a warm subprocess +
+`--resume <last_session_id>` on later cold spawns and crash recovery —
+all on a stable `cwd=repo_root`. (An interim `--continue` variant was
+reverted: the CLI resolves `--continue` only against interactive
+sessions, so it never found the pool's non-interactive ones.) The entry
+point is
[`claude_code_agent/__init__.py::execute_op`](../server/nodes/agent/claude_code_agent/__init__.py)
(sets `continue_session = bool(memory_data)`) and the warm-subprocess
pool at [`claude_code_agent/_pool.py`](../server/nodes/agent/claude_code_agent/_pool.py).
@@ -549,20 +552,20 @@ runs (Closes I-6).** Three coupled mechanisms:
skips `git worktree add`; `cleanup` skips `git worktree remove`.
Confirms claude's `project_key` is constant across runs.
-2. **`--continue` first-run / warm-subprocess multi-turn / `--resume`
- crash recovery.**
+2. **`--session-id` first-run / warm-subprocess multi-turn / `--resume`
+ later runs and crash recovery.**
[`claude_code_agent/__init__.py`](../server/nodes/agent/claude_code_agent/__init__.py)
- sets `continue_session = bool(memory_data)`; the argv-builder
+ sets `resume_session_id` from the memory node's persisted
+ `last_session_id`; the argv-builder
[`_provider.py::interactive_argv`](../server/nodes/agent/claude_code_agent/_provider.py)
- emits `--continue` on the first cold spawn for a memory-wired run
- (claude auto-loads the most recent conversation under the stable
- cwd's `project_key`). Subsequent turns are written as stream-json to
- the same warm subprocess held by
+ emits `--resume <UUID>` when present, otherwise the pool mints a
+ `uuid4` and emits `--session-id <UUID>` on the cold spawn. Subsequent
+ turns are written as stream-json to the same warm subprocess held by
[`_pool.py`](../server/nodes/agent/claude_code_agent/_pool.py), which
preserves the session UUID in-process. If the subprocess dies between
batches, the pool respawns with `--resume <current_session_uuid>`.
- `--continue` and `--resume` are mutually exclusive in argv emission.
- (This supersedes the earlier UUID5 / `--session-id` first-run dance.)
+ `--resume`, `--continue` and `--session-id` are mutually exclusive in
+ argv emission, in that precedence order.
3. **Auto-clear stale `last_session_id`.**
[`_persist_memory`](../server/services/cli_agent/service.py)
@@ -587,8 +590,8 @@ runs (Closes I-6).** Three coupled mechanisms:
origin; locked by `tests/test_cloudevents_node_parameters.py`.
5. **Parallel-batch guard.** `len(tasks) > 1` with memory wired raises
- `NodeUserError` at handler entry — concurrent `--continue` spawns
- against one project_key would race claude's session resolution.
+ `NodeUserError` at handler entry — concurrent `--resume` spawns
+ against one session would race claude's transcript writes.
Markdown is the UI mirror. The `simpleMemory.memory_content` field
keeps mirroring conversation turns via
diff --git a/docs-internal/cli_agent_framework.md b/docs-internal/cli_agent_framework.md
index e818fbb3..cac7bd09 100644
--- a/docs-internal/cli_agent_framework.md
+++ b/docs-internal/cli_agent_framework.md
@@ -124,15 +124,23 @@ back-compat; they're silently dropped in `interactive_argv`.
All flags documented at
[code.claude.com/docs/en/cli-reference](https://code.claude.com/docs/en/cli-reference).
-Worktree, lockfile, and bearer-token MCP server are wired in
-`nodes/agent/claude_code_agent/_pool.py:_spawn` (pool path) and
-`services/cli_agent/session.py:_pre_spawn` (non-pool path); the `--ide`
-flag tells the CLI to discover that lockfile via `CLAUDE_IDE_LOCK`.
-
-The non-pooled `AICliSession` path (one-shot prompt-in-argv runs that
-don't need session reuse) still uses PTY (`ptyprocess` POSIX /
-`pywinpty>=3.0.3` Windows) and remains out of scope for the
-subprocess+stream-json refactor.
+Skill materialisation and the bearer-token MCP server are wired in
+`nodes/agent/claude_code_agent/_pool.py:_spawn`. **Every Claude run
+routes through the pool** (`AICliService.run_batch`): memory- or
+Context-bound runs get one warm session keyed by the conversation;
+unbound runs get one ephemeral session per task keyed by
+`<node_id>:<turn_execution_id>:<index>` and terminated once the batch
+settles. `AICliSession` (PTY + on-disk JSONL watcher) is no longer used
+for Claude — a PTY stdin makes the CLI reject stream-json input, and
+that path never wrote the prompt (GitHub issues #133 / #134).
+
+Session identity follows the CLI reference: a cold spawn mints
+`--session-id <uuid4>` so the UUID is known before the first event;
+memory-bound runs pass `--resume <last_session_id>` (persisted on the
+memory node by `_persist_memory`). `--continue` is not used because the
+CLI skips non-interactively created sessions when resolving it. A child
+that exits without a `result` event wakes `send_turn` on stdout EOF and
+reports the exit code instead of burning the turn timeout.
Factory — a registry lookup, not a hardcoded branch per provider. Each plugin
calls `register_provider(name, factory)` on import (claude from
@@ -228,7 +236,7 @@ Steps:
1. Run `claude auth status`. If it exits 0, write the marker + broadcast and return immediately (idempotent re-click).
2. Otherwise schedule `_finalize_claude_login()` (in `nodes/agent/claude_code_agent/_handlers.py`), which calls `run_claude_login()` from `_oauth.py`:
- - OpenCompany-managed install of `@anthropic-ai/claude-code` into the shared npm tree at `<DATA_DIR>/packages/` via `npm install --prefix <packages_dir>` (same tree as `edgymeow` / `agent-browser`; skipped if already installed). Binary resolves to `<DATA_DIR>/packages/node_modules/.bin/claude[.cmd]`.
+ - OpenCompany-managed install of `@anthropic-ai/claude-code` into the shared npm tree at `<DATA_DIR>/packages/` via `npm install <package_name>@<package_version> --prefix <packages_dir>` (same tree as `edgymeow` / `agent-browser`; skipped if already installed). The exact version is pinned in `server/config/ai_cli_providers.json` (`package_version`, the vercel / cloudflare idiom) because the stream-json contract and flag surface are verified against that version only. Binary resolves to `<DATA_DIR>/packages/node_modules/.bin/claude[.cmd]`.
- `claude auth login` via `run_cli_command(..., env={..., CLAUDE_CONFIG_DIR=<DATA_DIR>/claude/}, stdin=asyncio.subprocess.PIPE)` — same way the VSCode Claude Code extension delegates to the binary. Anthropic doesn't expose `--print-url` or a programmatic OAuth helper (issue [anthropics/claude-code#7100](https://github.com/anthropics/claude-code/issues/7100), closed "not planned"), so we let the CLI open the user's browser via its own OS-level call. `stdin=PIPE` is **load-bearing** for claude-code >= 2.1.162's native binary: it reads stdin while waiting for the browser callback, and an inherited (closed) stdin EOFs it into an early exit that kills the localhost callback server before the redirect arrives — `stdin=PIPE` (never written) makes the read block so the server stays up.
3. Schedule a background task that polls `claude auth status` every 2s up to 600s. On exit-0, write the synthetic `"cli-managed"` marker via `auth_service.store_oauth_tokens("claude_code", ...)` and broadcast `credential_catalogue_updated`. The catalogue's `stored` flag flips and the existing `OAuthConnect.tsx` primitive renders the modal as Connected.
@@ -291,13 +299,13 @@ Shared by every CLI provider plugin. Imports nothing from `nodes/`.
| `mcp_server.py` | FastMCP sub-app mounted at `/mcp/ide` (JSON-RPC endpoint `/mcp/ide/mcp`) with bearer-token middleware + 7 infrastructure tools (`getWorkspaceFiles`, `listSkills`, `getSkill`, `readSkillResource`, `searchSkillResource`, `getCredential`, `broadcastLog`) + `rebind_batch` for warm-reuse context updates. |
| `context_bridge.py` | `SpecializedAgentContextBridge` — RFC-0002 Context continuity for specialized providers (`resolve` / `augment_prompt` / `record_turn`); shared with RLM and Vertex. |
| `workflow_tools.py` | Per-batch MCP tool exposure (`mcp__opencompany__<node_type>`) + handler scope check + `tools/list_changed` notify. |
-| `session.py` | `AICliSession(BaseProcessSupervisor)` — generic non-pool path (still PTY on POSIX). |
-| `service.py` | `AICliService.run_batch()` — dispatcher; routes to pool when memory-bound via `factory.get_session_pool(provider_name)`. |
+| `session.py` | `AICliSession(BaseProcessSupervisor)` — generic PTY + JSONL path for non-claude providers; not used for claude. |
+| `service.py` | `AICliService.run_batch()` — dispatcher; every claude task goes to the pool via `factory.get_session_pool(provider_name)` (warm keyed session when bound, ephemeral per-task session otherwise). |
| `_cli_auth.py` | CLI-agnostic `mark_logged_in` / `mark_logged_out` / `broadcast_credential_event` + `"cli-managed"` marker token. Shared by claude + codex handlers. |
| `_handlers.py` | Codex WS handlers (`codex_cli_login` / `codex_cli_logout`). Claude's moved to the plugin folder. |
| `providers/openai_codex.py` | Codex provider — sandbox-first, no session continuity. Will move when codex_agent adopts the per-folder layout. |
| `providers/google_gemini.py` | v2 stub. |
-| `jsonl_watcher.py` | Still used by the non-pooled `AICliSession` (out-of-scope path). |
+| `jsonl_watcher.py` | Used only by `AICliSession`; no claude caller. |
### Claude plugin — `server/nodes/agent/claude_code_agent/`
@@ -398,7 +406,8 @@ finds its own JSONL.
| Run state | Argv emitted | Why |
|---|---|---|
-| First cold spawn under a memory-wired node | `--continue` | Claude auto-loads the most recent conversation under cwd's `project_key` (per [code.claude.com/docs/en/cli-reference](https://code.claude.com/docs/en/cli-reference)). No UUID round-trip needed — claude tracks its own latest session per cwd on disk. |
+| First cold spawn, no stored session | `--session-id <uuid4>` | `ClaudeSessionPool._spawn` mints the UUID so it is known before the first event; `_persist_memory` stores it on the memory node as `last_session_id`. |
+| Cold spawn under a memory-wired node with a stored session | `--resume <last_session_id>` | Resume by UUID, never `--continue`: per [code.claude.com/docs/en/cli-reference](https://code.claude.com/docs/en/cli-reference) `--continue` skips sessions created non-interactively (`-p` / stream-json), which is every session the pool creates. |
| Subsequent turn, SAME warm subprocess | nothing argv-level — stream-json line on `proc.stdin` | The pool keeps the subprocess alive between turns. Claude maintains the conversation in-process; same `session_id` across turns (verified end-to-end). |
| Crash recovery (subprocess died between batches) | `--resume <captured_uuid>` | `ClaudeSessionPool.acquire` detects `process.returncode is not None`, captures the dead session's `current_session_uuid`, and respawns with `--resume`. Same `cwd=repo_root` → same `project_key` → claude finds the same JSONL it was writing before the crash. Mutually exclusive with `--continue`. |
@@ -438,7 +447,7 @@ ClaudeCodeAgentNode.execute_op (__init__.py:291-330)
│ connected_memory["node_id"]
│ ├─ cold: spawn `claude --output-format stream-json
│ │ --input-format stream-json --verbose --ide
- │ │ --continue ...` as subprocess with stdio pipes;
+ │ │ --session-id <uuid> | --resume <uuid> ...` as subprocess with stdio pipes;
│ │ no PTY. stdout_reader_task parses each event line
│ │ and dispatches to _handle_stream_event.
│ └─ warm reuse: return the existing PooledClaudeSession;
@@ -481,9 +490,9 @@ ClaudeCodeAgentNode.execute_op (__init__.py:291-330)
### Parallel-batch guard
-Memory continuity requires serial execution — N concurrent `--continue`
-spawns against the same `project_key` would race claude's
-session-resolution. When `memory_data` is wired AND `len(tasks) > 1`,
+Memory continuity requires serial execution — N concurrent `--resume`
+spawns against the same session would race claude's transcript
+writes. When `memory_data` is wired AND `len(tasks) > 1`,
`claude_code_agent` raises `NodeUserError("Memory-bound batches must
run one task at a time. ...")` at handler entry.
@@ -491,7 +500,8 @@ run one task at a time. ...")` at handler entry.
`memory_content` (the markdown surface the simpleMemory UI shows) is
a **display mirror**, not the resume channel. Claude's own JSONL on
-disk is what `--continue` and `--resume` load from. `_persist_memory`
+disk is what `--resume` loads from; `last_session_id` on the memory
+node is the resume handle. `_persist_memory`
appends each successful run's prompt + response to `memory_content`
via `append_to_memory_markdown` so the UI shows the conversation grow
live. User edits to `memory_content` do NOT influence claude's next
@@ -501,19 +511,19 @@ To reset both: click the simpleMemory's clear button or invoke the
`clear_memory` WS handler — backend wipes `memory_content` to the
default placeholder AND clears `last_session_id` in one DB write.
Claude's on-disk JSONL is left alone (orphan, harmless). The next run
-spawns fresh with `--continue` against a project_key that has no
-matching prior session, and claude assigns a brand-new UUID.
+has no `last_session_id`, so the pool mints a new `--session-id` and
+claude starts a brand-new session.
### Logs to watch
```
-[Claude Code memory] memory_node=<id> -> --continue
- (claude auto-finds latest session under cwd)
+[Claude Code memory] memory_node=<id> -> --resume <UUID>
+ (or "fresh session (no last_session_id yet)" on the first run)
[CC-Agent run_batch] enter ... memory=<memory_node_id> ...
[ClaudeSessionPool] spawned new session memory_node=<id> pid=<N>
- (cold spawn — argv carries --continue or --resume <UUID>)
+ (cold spawn — argv carries --session-id <UUID> or --resume <UUID>)
[ClaudeSessionPool] warm reuse memory_node=<id> pid=<N> uuid=<UUID>
(intra-process turn — stream-json on stdin)
diff --git a/docs-internal/errors.md b/docs-internal/errors.md
index 0468ec31..04b1e953 100644
--- a/docs-internal/errors.md
+++ b/docs-internal/errors.md
@@ -407,7 +407,7 @@ python -c "import ntpath; print(ntpath.join(r'D:\ws\AI_Employee_1', '2:processMa
---
-## 14. `npm install -g` Fails With `externally-managed-environment` (Ubuntu 24.04+)
+## 15. `npm install -g` Fails With `externally-managed-environment` (Ubuntu 24.04+)
**Symptom**: `npm install -g @zeenie-ai/opencompany` aborts inside the postinstall with:
```
diff --git a/docs-internal/node-logic-flows/specialized_agents/claude_code_agent.md b/docs-internal/node-logic-flows/specialized_agents/claude_code_agent.md
index b3bab56b..b9514659 100644
--- a/docs-internal/node-logic-flows/specialized_agents/claude_code_agent.md
+++ b/docs-internal/node-logic-flows/specialized_agents/claude_code_agent.md
@@ -16,8 +16,10 @@ as a plain subprocess driven over the VSCode-extension stream-json protocol
(`--output-format stream-json --input-format stream-json --verbose --ide`) —
NOT the old headless `claude -p`. The common case is a single visible
`prompt` that synthesises a one-task batch. Memory continuity is handled by
-claude's own on-disk session JSONL via `--continue` / `--resume` over a
-stable `cwd`, not by re-injecting markdown. This is the only
+claude's own on-disk session JSONL via a host-minted `--session-id` on the
+first run and `--resume <last_session_id>` afterwards (never `--continue`,
+which the CLI resolves only against interactive sessions) over a stable
+`cwd`, not by re-injecting markdown. This is the only
specialized-agent node that shells out; the others stay in-process.
## Inputs (handles)
@@ -28,7 +30,7 @@ Standard `std_agent_handles()` topology (same as the generic agents).
|--------|---------|
| `input-main` | Auto-prompt fallback (reads `source.message / text / content / str`) |
| `input-skill` | Connected skill names collected; SKILL.md trees materialised under `<workspace>/.claude/skills/` and exposed to the CLI (not injected into a system prompt) |
-| `input-memory` | When wired, sets `continue_session=True` so claude emits `--continue` and resumes its on-disk session under the stable cwd |
+| `input-memory` | When wired, sets `resume_session_id` from the memory node's persisted `last_session_id` so claude emits `--resume <UUID>` and continues its on-disk session; the first run has no UUID yet and the pool mints one via `--session-id` |
| `input-tools` | Connected tool nodes exposed to the CLI as `mcp__opencompany__<type>` MCP tools |
| `input-task` | Collected by the edge-walker; `task_data` is unpacked but the 5th element is ignored (`_`) here |
@@ -117,11 +119,13 @@ flowchart TD
`effort` + `fallback_model` + `continue_session`.
- **Batch defaulting**: when `params.tasks` is non-empty, node-level
`model`/`system_prompt`/`effort`/`fallback_model` fill any task that
- didn't override; `continue_session=True` is auto-set only when memory is
- wired AND the task didn't explicitly opt in or pick a resume UUID.
-- **Memory bridge**: `continue_session = bool(memory_data)`; the argv builder
- emits `--continue` (first run is a benign no-op, later runs resume the
- on-disk JSONL). NO markdown re-injection.
+ didn't override; `resume_session_id` is auto-set from the memory node's
+ `last_session_id` only when memory is wired AND the task didn't explicitly
+ opt in/out or pick its own UUID.
+- **Memory bridge**: `resume_session_id = memory_data["last_session_id"]`;
+ the argv builder emits `--resume <UUID>` (first run has none, so the pool
+ mints `--session-id` and `_persist_memory` stores it for next time). NO
+ markdown re-injection.
- **Serial-memory guard**: memory wired AND `len(tasks) > 1` -> `NodeUserError`
(parallel `--resume` against one JSONL would corrupt it).
- **Workspace**: `ctx.raw["workspace_dir"]` (injected by workflow.py) or
diff --git a/server/config/model_registry.json b/server/config/model_registry.json
index 810f2164..23e1815b 100644
--- a/server/config/model_registry.json
+++ b/server/config/model_registry.json
@@ -1,8 +1,101 @@
{
- "_generated": "2026-09-05T05:04:06.930651+00:00",
+ "_generated": "2026-09-09T07:22:39.821023+00:00",
"_source": "openrouter",
"_model_count": 431,
"models": {
+ "inception/mercury-2.5": {
+ "id": "inception/mercury-2.5",
+ "name": "Inception: Mercury 2.5",
+ "provider": "inception",
+ "local_id": "mercury-2.5",
+ "context_length": 260000,
+ "max_output_tokens": 65536,
+ "input_price_per_mtok": 0.04,
+ "output_price_per_mtok": 0.15,
+ "supports_thinking": false,
+ "thinking_type": "none",
+ "temperature_range": [
+ 0.0,
+ 2.0
+ ],
+ "is_reasoning_model": false,
+ "supported_parameters": [
+ "include_reasoning",
+ "max_tokens",
+ "reasoning",
+ "reasoning_effort",
+ "response_format",
+ "stop",
+ "structured_outputs",
+ "temperature",
+ "tool_choice",
+ "tools"
+ ]
+ },
+ "nex-agi/nex-n2.5-mini:free": {
+ "id": "nex-agi/nex-n2.5-mini:free",
+ "name": "Nex AGI: Nex-N2.5-Mini (free)",
+ "provider": "nex-agi",
+ "local_id": "nex-n2.5-mini:free",
+ "context_length": 262144,
+ "max_output_tokens": 235929,
+ "input_price_per_mtok": 0.0,
+ "output_price_per_mtok": 0.0,
+ "supports_thinking": false,
+ "thinking_type": "none",
+ "temperature_range": [
+ 0.0,
+ 2.0
+ ],
+ "is_reasoning_model": false,
+ "supported_parameters": [
+ "include_reasoning",
+ "logprobs",
+ "max_tokens",
+ "reasoning",
+ "reasoning_effort",
+ "response_format",
+ "structured_outputs",
+ "temperature",
+ "tool_choice",
+ "tools",
+ "top_k",
+ "top_logprobs",
+ "top_p"
+ ]
+ },
+ "nex-agi/nex-n2.5-pro:free": {
+ "id": "nex-agi/nex-n2.5-pro:free",
+ "name": "Nex AGI: Nex-N2.5-Pro (free)",
+ "provider": "nex-agi",
+ "local_id": "nex-n2.5-pro:free",
+ "context_length": 262144,
+ "max_output_tokens": 235929,
+ "input_price_per_mtok": 0.0,
+ "output_price_per_mtok": 0.0,
+ "supports_thinking": false,
+ "thinking_type": "none",
+ "temperature_range": [
+ 0.0,
+ 2.0
+ ],
+ "is_reasoning_model": false,
+ "supported_parameters": [
+ "include_reasoning",
+ "logprobs",
+ "max_tokens",
+ "reasoning",
+ "reasoning_effort",
+ "response_format",
+ "structured_outputs",
+ "temperature",
+ "tool_choice",
+ "tools",
+ "top_k",
+ "top_logprobs",
+ "top_p"
+ ]
+ },
"openai/gpt-6-astra": {
"id": "openai/gpt-6-astra",
"name": "OpenAI: GPT-6 Astra",
@@ -367,35 +460,6 @@
"verbosity"
]
},
- "inception/mercury-2.5-preview": {
- "id": "inception/mercury-2.5-preview",
- "name": "Inception: Mercury 2.5 Preview",
- "provider": "inception",
- "local_id": "mercury-2.5-preview",
- "context_length": 260000,
- "max_output_tokens": 65536,
- "input_price_per_mtok": 0.04,
- "output_price_per_mtok": 0.15,
- "supports_thinking": false,
- "thinking_type": "none",
- "temperature_range": [
- 0.0,
- 2.0
- ],
- "is_reasoning_model": false,
- "supported_parameters": [
- "include_reasoning",
- "max_tokens",
- "reasoning",
- "reasoning_effort",
- "response_format",
- "stop",
- "structured_outputs",
- "temperature",
- "tool_choice",
- "tools"
- ]
- },
"ibm-granite/granite-4.2-8b": {
"id": "ibm-granite/granite-4.2-8b",
"name": "IBM: Granite 4.2 8B",
@@ -403,8 +467,8 @@
"local_id": "granite-4.2-8b",
"context_length": 131072,
"max_output_tokens": 117964,
- "input_price_per_mtok": 0.1,
- "output_price_per_mtok": 0.15,
+ "input_price_per_mtok": 0.06,
+ "output_price_per_mtok": 0.25,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -653,10 +717,10 @@
"name": "Z.ai: GLM 5.3 Flash (batch)",
"provider": "z-ai",
"local_id": "glm-5.3-flash:batch",
- "context_length": 1048575,
- "max_output_tokens": 943717,
- "input_price_per_mtok": 0.15,
- "output_price_per_mtok": 0.5,
+ "context_length": 1048576,
+ "max_output_tokens": 943718,
+ "input_price_per_mtok": 0.075,
+ "output_price_per_mtok": 0.25,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -723,7 +787,7 @@
"provider": "deepseek",
"local_id": "deepseek-v4-flash-vision-exp",
"context_length": 1048576,
- "max_output_tokens": 384000,
+ "max_output_tokens": 943718,
"input_price_per_mtok": 0.22,
"output_price_per_mtok": 0.66,
"supports_thinking": false,
@@ -756,6 +820,43 @@
"top_p"
]
},
+ "deepseek/deepseek-v4-flash-vision-exp:batch": {
+ "id": "deepseek/deepseek-v4-flash-vision-exp:batch",
+ "name": "DeepSeek: DeepSeek V4 Flash Vision Exp (batch)",
+ "provider": "deepseek",
+ "local_id": "deepseek-v4-flash-vision-exp:batch",
+ "context_length": 1048576,
+ "max_output_tokens": 943718,
+ "input_price_per_mtok": 0.11,
+ "output_price_per_mtok": 0.33,
+ "supports_thinking": false,
+ "thinking_type": "none",
+ "temperature_range": [
+ 0.0,
+ 2.0
+ ],
+ "is_reasoning_model": false,
+ "supported_parameters": [
+ "frequency_penalty",
+ "include_reasoning",
+ "logit_bias",
+ "logprobs",
+ "max_tokens",
+ "presence_penalty",
+ "reasoning",
+ "reasoning_effort",
+ "repetition_penalty",
+ "response_format",
+ "stop",
+ "structured_outputs",
+ "temperature",
+ "tool_choice",
+ "tools",
+ "top_k",
+ "top_logprobs",
+ "top_p"
+ ]
+ },
"tencent/hy-mt2-1.8b": {
"id": "tencent/hy-mt2-1.8b",
"name": "Tencent: Hy-MT2-1.8B",
@@ -810,9 +911,9 @@
"provider": "z-ai",
"local_id": "glm-latest",
"context_length": 1310720,
- "max_output_tokens": 235929,
- "input_price_per_mtok": 1.15,
- "output_price_per_mtok": 3.5,
+ "max_output_tokens": 128000,
+ "input_price_per_mtok": 1.113,
+ "output_price_per_mtok": 3.498,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -875,7 +976,7 @@
"provider": "z-ai",
"local_id": "glm-5.3",
"context_length": 1310720,
- "max_output_tokens": 262144,
+ "max_output_tokens": 943718,
"input_price_per_mtok": 1.4,
"output_price_per_mtok": 4.4,
"supports_thinking": false,
@@ -909,6 +1010,43 @@
"top_p"
]
},
+ "z-ai/glm-5.3:batch": {
+ "id": "z-ai/glm-5.3:batch",
+ "name": "Z.ai: GLM 5.3 (batch)",
+ "provider": "z-ai",
+ "local_id": "glm-5.3:batch",
+ "context_length": 1048576,
+ "max_output_tokens": 943718,
+ "input_price_per_mtok": 0.7,
+ "output_price_per_mtok": 2.2,
+ "supports_thinking": false,
+ "thinking_type": "none",
+ "temperature_range": [
+ 0.0,
+ 2.0
+ ],
+ "is_reasoning_model": false,
+ "supported_parameters": [
+ "frequency_penalty",
+ "include_reasoning",
+ "logit_bias",
+ "logprobs",
+ "max_tokens",
+ "presence_penalty",
+ "reasoning",
+ "reasoning_effort",
+ "repetition_penalty",
+ "response_format",
+ "stop",
+ "structured_outputs",
+ "temperature",
+ "tool_choice",
+ "tools",
+ "top_k",
+ "top_logprobs",
+ "top_p"
+ ]
+ },
"qwen/qwen3.8-27b": {
"id": "qwen/qwen3.8-27b",
"name": "Qwen: Qwen3.8 27B",
@@ -1072,7 +1210,7 @@
"provider": "qwen",
"local_id": "qwen3.8-2.4t-a95b",
"context_length": 1048576,
- "max_output_tokens": 262144,
+ "max_output_tokens": 131072,
"input_price_per_mtok": 2.0,
"output_price_per_mtok": 6.0,
"supports_thinking": false,
@@ -1180,9 +1318,9 @@
"provider": "deepseek",
"local_id": "deepseek-v4-pro-0813",
"context_length": 1048576,
- "max_output_tokens": 384000,
- "input_price_per_mtok": 1.1207,
- "output_price_per_mtok": 3.362,
+ "max_output_tokens": 393216,
+ "input_price_per_mtok": 0.5795,
+ "output_price_per_mtok": 1.7384,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -1220,8 +1358,8 @@
"local_id": "deepseek-v4-pro-0813:batch",
"context_length": 1048576,
"max_output_tokens": 943718,
- "input_price_per_mtok": 1.32,
- "output_price_per_mtok": 3.96,
+ "input_price_per_mtok": 0.66,
+ "output_price_per_mtok": 1.98,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -1233,6 +1371,7 @@
"frequency_penalty",
"include_reasoning",
"logit_bias",
+ "logprobs",
"max_tokens",
"min_p",
"presence_penalty",
@@ -1246,6 +1385,7 @@
"tool_choice",
"tools",
"top_k",
+ "top_logprobs",
"top_p"
]
},
@@ -1489,8 +1629,8 @@
"local_id": "muse-glimmer-30b:batch",
"context_length": 131072,
"max_output_tokens": 117964,
- "input_price_per_mtok": 0.35,
- "output_price_per_mtok": 1.5,
+ "input_price_per_mtok": 0.175,
+ "output_price_per_mtok": 0.75,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -1502,6 +1642,7 @@
"frequency_penalty",
"include_reasoning",
"logit_bias",
+ "logprobs",
"max_tokens",
"min_p",
"presence_penalty",
@@ -1515,6 +1656,7 @@
"tool_choice",
"tools",
"top_k",
+ "top_logprobs",
"top_p"
]
},
@@ -1555,9 +1697,9 @@
"provider": "deepseek",
"local_id": "deepseek-v4-flash-latest",
"context_length": 1310720,
- "max_output_tokens": 131072,
+ "max_output_tokens": 393216,
"input_price_per_mtok": 0.05,
- "output_price_per_mtok": 0.1,
+ "output_price_per_mtok": 0.16,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -1638,8 +1780,8 @@
"local_id": "deepseek-v4-flash-0731:batch",
"context_length": 1048576,
"max_output_tokens": 943718,
- "input_price_per_mtok": 0.14,
- "output_price_per_mtok": 0.28,
+ "input_price_per_mtok": 0.11,
+ "output_price_per_mtok": 0.33,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -1651,6 +1793,7 @@
"frequency_penalty",
"include_reasoning",
"logit_bias",
+ "logprobs",
"max_tokens",
"min_p",
"presence_penalty",
@@ -1664,6 +1807,7 @@
"tool_choice",
"tools",
"top_k",
+ "top_logprobs",
"top_p"
]
},
@@ -1759,6 +1903,7 @@
"supported_parameters": [
"frequency_penalty",
"include_reasoning",
+ "logprobs",
"max_tokens",
"presence_penalty",
"reasoning",
@@ -1767,6 +1912,7 @@
"stop",
"temperature",
"tools",
+ "top_logprobs",
"top_p"
]
},
@@ -3020,37 +3166,6 @@
"top_p"
]
},
- "nex-agi/nex-n2-mini": {
- "id": "nex-agi/nex-n2-mini",
- "name": "Nex AGI: Nex-N2-Mini",
- "provider": "nex-agi",
- "local_id": "nex-n2-mini",
- "context_length": 262144,
- "max_output_tokens": 235929,
- "input_price_per_mtok": 0.025,
- "output_price_per_mtok": 0.1,
- "supports_thinking": false,
- "thinking_type": "none",
- "temperature_range": [
- 0.0,
- 2.0
- ],
- "is_reasoning_model": false,
- "supported_parameters": [
- "include_reasoning",
- "logprobs",
- "max_tokens",
- "reasoning",
- "response_format",
- "structured_outputs",
- "temperature",
- "tool_choice",
- "tools",
- "top_k",
- "top_logprobs",
- "top_p"
- ]
- },
"sakana/fugu-ultra": {
"id": "sakana/fugu-ultra",
"name": "Sakana: Fugu Ultra",
@@ -3206,15 +3321,15 @@
"top_p"
]
},
- "z-ai/glm-5.2:free": {
- "id": "z-ai/glm-5.2:free",
- "name": "Z.ai: GLM 5.2 (free)",
+ "z-ai/glm-5.2:batch": {
+ "id": "z-ai/glm-5.2:batch",
+ "name": "Z.ai: GLM 5.2 (batch)",
"provider": "z-ai",
- "local_id": "glm-5.2:free",
- "context_length": 256000,
- "max_output_tokens": 230400,
- "input_price_per_mtok": 0.0,
- "output_price_per_mtok": 0.0,
+ "local_id": "glm-5.2:batch",
+ "context_length": 1048576,
+ "max_output_tokens": 943718,
+ "input_price_per_mtok": 0.7,
+ "output_price_per_mtok": 2.2,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -3225,20 +3340,21 @@
"supported_parameters": [
"frequency_penalty",
"include_reasoning",
+ "logit_bias",
+ "logprobs",
"max_tokens",
- "min_p",
"presence_penalty",
"reasoning",
"reasoning_effort",
"repetition_penalty",
"response_format",
- "seed",
"stop",
"structured_outputs",
"temperature",
"tool_choice",
"tools",
"top_k",
+ "top_logprobs",
"top_p"
]
},
@@ -3267,8 +3383,8 @@
"local_id": "kimi-k2.7-code",
"context_length": 262144,
"max_output_tokens": 235929,
- "input_price_per_mtok": 0.66,
- "output_price_per_mtok": 3.4,
+ "input_price_per_mtok": 0.71,
+ "output_price_per_mtok": 3.5,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -3387,36 +3503,6 @@
"verbosity"
]
},
- "nex-agi/nex-n2-pro": {
- "id": "nex-agi/nex-n2-pro",
- "name": "Nex AGI: Nex-N2-Pro",
- "provider": "nex-agi",
- "local_id": "nex-n2-pro",
- "context_length": 262144,
- "max_output_tokens": 235929,
- "input_price_per_mtok": 0.25,
- "output_price_per_mtok": 1.0,
- "supports_thinking": false,
- "thinking_type": "none",
- "temperature_range": [
- 0.0,
- 2.0
- ],
- "is_reasoning_model": false,
- "supported_parameters": [
- "frequency_penalty",
- "include_reasoning",
- "logprobs",
- "max_tokens",
- "reasoning",
- "temperature",
- "tool_choice",
- "tools",
- "top_k",
- "top_logprobs",
- "top_p"
- ]
- },
"nvidia/nemotron-3.5-content-safety": {
"id": "nvidia/nemotron-3.5-content-safety",
"name": "NVIDIA: Nemotron 3.5 Content Safety",
@@ -3648,34 +3734,6 @@
"top_p"
]
},
- "minimax/minimax-m3:free": {
- "id": "minimax/minimax-m3:free",
- "name": "MiniMax: MiniMax M3 (free)",
- "provider": "minimax",
- "local_id": "minimax-m3:free",
- "context_length": 1048576,
- "max_output_tokens": 943718,
- "input_price_per_mtok": 0.0,
- "output_price_per_mtok": 0.0,
- "supports_thinking": false,
- "thinking_type": "none",
- "temperature_range": [
- 0.0,
- 2.0
- ],
- "is_reasoning_model": false,
- "supported_parameters": [
- "include_reasoning",
- "max_tokens",
- "reasoning",
- "response_format",
- "seed",
- "temperature",
- "tool_choice",
- "tools",
- "top_p"
- ]
- },
"stepfun/step-3.7-flash": {
"id": "stepfun/step-3.7-flash",
"name": "StepFun: Step 3.7 Flash",
@@ -4120,8 +4178,8 @@
"name": "Mistral: Mistral Medium 3.5 (batch)",
"provider": "mistralai",
"local_id": "mistral-medium-3-5:batch",
- "context_length": 32768,
- "max_output_tokens": 26214,
+ "context_length": 262144,
+ "max_output_tokens": 209715,
"input_price_per_mtok": 0.75,
"output_price_per_mtok": 3.75,
"supports_thinking": false,
@@ -4273,8 +4331,8 @@
"local_id": "kimi-latest",
"context_length": 1048576,
"max_output_tokens": 943718,
- "input_price_per_mtok": 2.5,
- "output_price_per_mtok": 14.0,
+ "input_price_per_mtok": 2.4,
+ "output_price_per_mtok": 12.0,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -4696,8 +4754,8 @@
"local_id": "deepseek-v4-pro",
"context_length": 1048576,
"max_output_tokens": 384000,
- "input_price_per_mtok": 0.9613,
- "output_price_per_mtok": 1.9227,
+ "input_price_per_mtok": 0.9553,
+ "output_price_per_mtok": 1.9105,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -4736,8 +4794,8 @@
"local_id": "deepseek-v4-flash",
"context_length": 1048576,
"max_output_tokens": 384000,
- "input_price_per_mtok": 0.0867,
- "output_price_per_mtok": 0.1733,
+ "input_price_per_mtok": 0.0886,
+ "output_price_per_mtok": 0.1772,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -5559,34 +5617,6 @@
"top_p"
]
},
- "minimax/minimax-m2.7:free": {
- "id": "minimax/minimax-m2.7:free",
- "name": "MiniMax: MiniMax M2.7 (free)",
- "provider": "minimax",
- "local_id": "minimax-m2.7:free",
- "context_length": 196608,
- "max_output_tokens": 176947,
- "input_price_per_mtok": 0.0,
- "output_price_per_mtok": 0.0,
- "supports_thinking": false,
- "thinking_type": "none",
- "temperature_range": [
- 0.0,
- 2.0
- ],
- "is_reasoning_model": false,
- "supported_parameters": [
- "include_reasoning",
- "max_tokens",
- "reasoning",
- "response_format",
- "seed",
- "temperature",
- "tool_choice",
- "tools",
- "top_p"
- ]
- },
"openai/gpt-5.4-nano": {
"id": "openai/gpt-5.4-nano",
"name": "OpenAI: GPT-5.4 Nano",
@@ -5768,7 +5798,7 @@
"name": "NVIDIA: Nemotron 3 Super",
"provider": "nvidia",
"local_id": "nemotron-3-super-120b-a12b",
- "context_length": 1000000,
+ "context_length": 262144,
"max_output_tokens": 16384,
"input_price_per_mtok": 0.085,
"output_price_per_mtok": 0.4,
@@ -5783,7 +5813,6 @@
"frequency_penalty",
"include_reasoning",
"logit_bias",
- "logprobs",
"max_tokens",
"min_p",
"presence_penalty",
@@ -5793,12 +5822,10 @@
"response_format",
"seed",
"stop",
- "structured_outputs",
"temperature",
"tool_choice",
"tools",
"top_k",
- "top_logprobs",
"top_p"
]
},
@@ -6175,8 +6202,8 @@
"local_id": "qwen3.5-35b-a3b",
"context_length": 262144,
"max_output_tokens": 16384,
- "input_price_per_mtok": 0.08,
- "output_price_per_mtok": 0.75,
+ "input_price_per_mtok": 0.3125,
+ "output_price_per_mtok": 1.25,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -7064,8 +7091,8 @@
"provider": "z-ai",
"local_id": "glm-4.7-flash",
"context_length": 202752,
- "max_output_tokens": 16384,
- "input_price_per_mtok": 0.06,
+ "max_output_tokens": 117964,
+ "input_price_per_mtok": 0.0605,
"output_price_per_mtok": 0.4,
"supports_thinking": false,
"thinking_type": "none",
@@ -8021,7 +8048,7 @@
"provider": "moonshotai",
"local_id": "kimi-k2-thinking",
"context_length": 262144,
- "max_output_tokens": 100352,
+ "max_output_tokens": 235929,
"input_price_per_mtok": 0.6,
"output_price_per_mtok": 2.5,
"supports_thinking": false,
@@ -8623,9 +8650,9 @@
"provider": "z-ai",
"local_id": "glm-4.6",
"context_length": 204800,
- "max_output_tokens": 131072,
- "input_price_per_mtok": 0.55,
- "output_price_per_mtok": 2.2,
+ "max_output_tokens": 16384,
+ "input_price_per_mtok": 0.43,
+ "output_price_per_mtok": 1.75,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -9019,7 +9046,7 @@
"provider": "qwen",
"local_id": "qwen3-next-80b-a3b-thinking",
"context_length": 262144,
- "max_output_tokens": 32768,
+ "max_output_tokens": 235929,
"input_price_per_mtok": 0.15,
"output_price_per_mtok": 1.2,
"supports_thinking": false,
@@ -9055,8 +9082,8 @@
"provider": "qwen",
"local_id": "qwen3-next-80b-a3b-instruct",
"context_length": 262144,
- "max_output_tokens": 235929,
- "input_price_per_mtok": 0.1,
+ "max_output_tokens": 16384,
+ "input_price_per_mtok": 0.09,
"output_price_per_mtok": 1.1,
"supports_thinking": false,
"thinking_type": "none",
@@ -9246,9 +9273,9 @@
"provider": "deepseek",
"local_id": "deepseek-chat-v3.1",
"context_length": 163840,
- "max_output_tokens": 144900,
- "input_price_per_mtok": 0.55,
- "output_price_per_mtok": 1.65,
+ "max_output_tokens": 32768,
+ "input_price_per_mtok": 0.25,
+ "output_price_per_mtok": 0.95,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -10047,8 +10074,8 @@
"local_id": "qwen3-235b-a22b-2507",
"context_length": 262144,
"max_output_tokens": 16384,
- "input_price_per_mtok": 0.09,
- "output_price_per_mtok": 0.55,
+ "input_price_per_mtok": 0.22,
+ "output_price_per_mtok": 0.88,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -10696,6 +10723,7 @@
"response_format",
"seed",
"stop",
+ "structured_outputs",
"temperature",
"tool_choice",
"tools",
@@ -10741,9 +10769,9 @@
"provider": "qwen",
"local_id": "qwen3-14b",
"context_length": 131072,
- "max_output_tokens": 16384,
- "input_price_per_mtok": 0.12,
- "output_price_per_mtok": 0.24,
+ "max_output_tokens": 8192,
+ "input_price_per_mtok": 0.2275,
+ "output_price_per_mtok": 0.91,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -11219,8 +11247,8 @@
"local_id": "deepseek-chat-v3-0324",
"context_length": 163840,
"max_output_tokens": 147456,
- "input_price_per_mtok": 0.25,
- "output_price_per_mtok": 1.0,
+ "input_price_per_mtok": 0.29,
+ "output_price_per_mtok": 1.14,
"supports_thinking": false,
"thinking_type": "none",
"temperature_range": [
@@ -12292,7 +12320,7 @@
"provider": "thedrummer",
"local_id": "unslopnemo-12b",
"context_length": 1024000,
- "max_output_tokens": 26214,
+ "max_output_tokens": 819200,
"input_price_per_mtok": 0.4,
"output_price_per_mtok": 0.4,
"supports_thinking": false,