Self-improving AI Employees for everyone
Python ai-agentsai-employeeai-employeesassistantautonomous-agentsco-employeescompanygrok-botinstinctloop-agentsmuseown-your-datapersonalself-hosted
docs: sweep the docs for the model refresh, the CodeQL fixes and stale counts
trohitg9fe30ec8c2a6
diff --git a/CLAUDE.md b/CLAUDE.mdindex e940f9d0..b6bad32c 100644--- a/CLAUDE.md+++ b/CLAUDE.md@@ -236,7 +236,7 @@ server/config/# services/plugin/capabilities.CapabilityConfig.server/nodejs/ # JS/TS executor sidecar — Express on bun (the only JS runtime shipped)-├── package.json # Dependency: express (inlined by `bun build --target=bun`); engines.bun >= 1.4.0+├── package.json # Dependencies: express + express-rate-limit (both inlined by `bun build --target=bun`); engines.bun >= 1.4.0├── tsconfig.json # TypeScript config (ES2024)└── src/└── index.ts # Express server (/execute, /health, /packages/*)@@ -268,7 +268,7 @@ Persistent Express server for JavaScript/TypeScript code execution, replacing su**Files:**```server/nodejs/-├── package.json # Dependency: express (inlined into the bundle). Scripts: build = `bun build src/index.ts --target=bun --outfile=dist/index.js`,+├── package.json # Dependencies: express + express-rate-limit (both inlined into the bundle; the limiter guards the /packages routes). Scripts: build = `bun build src/index.ts --target=bun --outfile=dist/index.js`,│ # start = `bun dist/index.js`, dev = `bun --watch src/index.ts`; engines = { bun >= 1.4.0 }. No esbuild, no tsx.├── tsconfig.json # TypeScript config (ES2024)├── dist/index.js # Self-contained bundle (gitignored; built by `company build` / the desktop stage script; ships pre-built in the tarball)@@ -285,8 +285,8 @@ server/nodes/code/ # Executor plugins (javascript_executor, types|----------|--------|-------------|| `/health` | GET | Health check: `runtime: "bun"`, `runtime_version`, and still `node_version` (the Node API level bun reports) || `/execute` | POST | Execute JS/TS code with input_data and timeout (`language: "typescript"` is type-stripped by `Bun.Transpiler` first, then both run in a `node:vm` context) |-| `/packages/install` | POST | Install npm-registry packages into user-packages via `bun add` (run through `process.execPath`, i.e. the sidecar's own bun) |-| `/packages` | GET | List installed packages from the user-packages dir's `package.json` |+| `/packages/install` | POST | Install npm-registry packages into user-packages via `bun add` (run through `process.execPath`, i.e. the sidecar's own bun). Rate-limited to 120 req/min |+| `/packages` | GET | List installed packages from the user-packages dir's `package.json`. Rate-limited to 120 req/min, sharing the window with `/packages/install` |**Environment Variables:**| Variable | Default | Description |@@ -2765,7 +2765,7 @@ This function:- **`outputMode: "terminal"` uiHint — spec-driven CLI output rendering**: the output panel (`components/output/OutputPanel.tsx`, the active renderer — `ui/OutputDisplayPanel.tsx` is legacy/unimported) renders string responses through ReactMarkdown by default, which whitespace-collapses CLI text. CLI-wrapper plugins (`githubAction`, `vercelAction`, `shell`) declare `ui_hints = {"outputMode": "terminal"}`; the panel resolves the spec via `useNodeSpec(selectedNode?.type)` and renders their text in a `<pre>` on the per-theme `--code-*` surface, routes wholly-JSON strings to the JSON tree via the shared `tryParseJson` (`utils/formatters.ts`), and surfaces object/array `result` payloads (server-side-parsed CLI JSON; arrays survive `unwrap` un-peeled) in the Response section. Pair with the `_shape` convention: parsed JSON in `result` OR text in `stdout` — never both — empty keys omitted. Locked by `client/src/components/__tests__/OutputPanel.test.tsx` (render tests: real `\n`/`\t` preserved) + the uiHints known-set in `test_node_spec.py`. Cache note: nodeSpecs are session-sticky (`staleTime: FOREVER`, revision-busted at page load) — after a backend restart a hard browser reload is needed before new uiHints reach an open canvas.- **`--action-X-hover` triplet + ActionButton zero-arithmetic**: each of the 6 action roles (`run`/`stop`/`save`/`config`/`secret`/`tools`) now exposes a `-hover` variant (0.25 alpha) alongside the existing `-soft` (0.15) and `-border` (0.6). ActionButton's CVA reads `hover:bg-action-X-hover` directly; disabled state is the shadcn-idiomatic `disabled:opacity-50` on the base class. No per-token `/25`, `/40`, `/10` opacity arithmetic at any call site. Credential-modal panels (`OAuthConnect`, `EmailPanel`, `QrPairingPanel`) and the skill editor (`MasterSkillEditor`) consume `<ActionButton intent="...">` directly. `ActionDef` carries an `intent` key; the catalogue adapter maps server-sent `theme_color` palette strings to intents via `SERVER_COLOR_TO_INTENT`.- **Credentials: DB as single source of truth + symmetric broadcasts + cache dedup**: `CredentialsDatabase` (encrypted SQLite) is canonical; every other layer is a derived cache with explicit invalidation. Backend `AuthService._memory_cache + _models_cache` collapsed into one `_api_key_cache: Dict[str, ApiKeyCacheEntry]` dataclass — single write/evict site. Per RFC 9700 (OAuth 2.0 BCP, 2024) the `_oauth_cache` no longer carries `refresh_token`; new `get_oauth_refresh_token(provider, customer)` reads from the encrypted DB on every call. `validate_api_key`, `save_api_key`, `delete_api_key`, `twitter_logout`, `google_logout` now broadcast symmetrically: `update_api_key_status` (in-memory map) + `broadcast_credential_event(...)` (refetch signal) wrapping `WorkflowEvent` (CloudEvents v1.0 from `services/events/envelope.py`, the same envelope the Wave 12 EventSource framework uses). The dead-letter `credential_catalogue_updated` event is finally emitted by the backend. Frontend retired the 200-LOC `client/src/components/credentials/providers.tsx` static fallback — `useCatalogueQuery` is the only source; cold-boot renders `<Skeleton>`, server-unreachable shows an explicit error state. `ApiKeyStatus.hasKey` mirror dropped (duplicated catalogue's `provider.stored`); two new selector hooks (`useProviderStored`, `useStoredProviderCount`) read the catalogue. Pytest invariant `test_credential_broadcasts.py` (17 tests) locks the broadcast contract via `inspect.getsource` introspection + the CloudEvents v1.0 envelope shape + AuthService DB-write-then-cache-update ordering + the no-refresh-token-in-cache rule.-- **Local-LLM provider routing + per-model context**: Ollama and LM Studio are first-class providers (12 total for agents: 10 cloud + 2 local). Provider detection is driven by `detect_ai_provider` in `server/constants.py` plus the `provider` Literal in `nodes/agent/{ai_agent,chat_agent,_specialized}.py` — both MUST list `ollama` / `lmstudio` or chat-model nodes / agent dropdowns silently fall through to `'openai'` and `execute_chat` calls api.openai.com with the placeholder key. The validator at `nodes/model/_local_validator.py` probes via the official SDKs (`ollama.AsyncClient.ps()` for typed `ProcessResponse.Model`, `lmstudio.AsyncClient.llm.list_loaded()` for typed `LlmInstanceInfo`) — reads only typed fields (`context_length`, `max_context_length`, `vision`, `trained_for_tool_use`, `architecture`, `params_string`, `format`, plus Ollama's `details.{family,parameter_size,quantization_level}`). No regex, no Modelfile-parameters parsing, no `/api/show` modelinfo dict-key hunting. Per-model params persist in `EncryptedAPIKey.models["model_params"]` (via `save_api_key(model_params=...)`) AND in `model_registry.json` via `register_local_model()` — sync `get_context_length()` / `get_max_output_tokens()` find real values without async DB lookups, and entries survive process restart. `is_model_valid_for_provider` returns `True` for open-world providers (`openrouter` / `ollama` / `lmstudio`) so local model names like `qwen/qwen3.6-27b` aren't rejected by the cloud-style pattern check. Runtime path uses `OpenAIProvider(base_url={user_proxy_url}, api_key="ollama")` — traffic stays on `localhost`.+- **Local-LLM provider routing + per-model context**: Ollama and LM Studio are first-class providers (13 total for agents: 11 cloud + 2 local). Provider detection is driven by `detect_ai_provider` in `server/constants.py` plus the `provider` Literal in `nodes/agent/{ai_agent,chat_agent,_specialized}.py` — both MUST list `ollama` / `lmstudio` or chat-model nodes / agent dropdowns silently fall through to `'openai'` and `execute_chat` calls api.openai.com with the placeholder key. The validator at `nodes/model/_local_validator.py` probes via the official SDKs (`ollama.AsyncClient.ps()` for typed `ProcessResponse.Model`, `lmstudio.AsyncClient.llm.list_loaded()` for typed `LlmInstanceInfo`) — reads only typed fields (`context_length`, `max_context_length`, `vision`, `trained_for_tool_use`, `architecture`, `params_string`, `format`, plus Ollama's `details.{family,parameter_size,quantization_level}`). No regex, no Modelfile-parameters parsing, no `/api/show` modelinfo dict-key hunting. Per-model params persist in `EncryptedAPIKey.models["model_params"]` (via `save_api_key(model_params=...)`) AND in `model_registry.json` via `register_local_model()` — sync `get_context_length()` / `get_max_output_tokens()` find real values without async DB lookups, and entries survive process restart. `is_model_valid_for_provider` returns `True` for open-world providers (`openrouter` / `ollama` / `lmstudio`) so local model names like `qwen/qwen3.6-27b` aren't rejected by the cloud-style pattern check. Runtime path uses `OpenAIProvider(base_url={user_proxy_url}, api_key="ollama")` — traffic stays on `localhost`.- **Typed SDK error → `LLMError` → `NodeUserError`**: Native providers normalize SDK failures into structured `LLMError` values carrying category, retryability, HTTP status, provider code, request ID, and retry-after metadata. `ChatUnifier` translates them at the execution boundary into a user-safe `NodeUserError`; `BaseNode.execute()` logs that at WARN with one line and no stack trace. Unexpected exceptions retain the full-traceback path. In-process agent turns retry only errors marked retryable, while Temporal owns activity retries and disables the inner retry loop.- **Plugin extraction (Wave 11.I)**: every plugin's WS handlers, OAuth client, FastAPI router, and lifecycle service live entirely under `nodes/<plugin>/`. Eight plugin domains migrated this round (whatsapp / twitter / google-workspace / android / browser / email / code / credential-validation-scaffold for maps+apify+ollama+lmstudio) follow the telegram pattern. `routers/websocket.py` shrunk from ~3,785 to ~2,977 LOC (-808). Three plugin routers (`routers/twitter.py`, `routers/google.py`, `routers/android.py`) moved into `nodes/<plugin>/_router.py` and mount via the plugin-router loop in `main.py`; the explicit `app.include_router(<plugin>.router)` calls and `from routers import <plugin>` imports in `main.py` are gone. **`register_router(router, name=...)` is the new sibling helper to `register_ws_handlers` (same file: `services/ws_handler_registry.py`)** — the generic registry set has since grown to 18 (see the Plugin System row at the top of this file). **`Credential.validate(data) -> dict` + `Credential._probe(api_key) -> ProbeResult` is the new shared validator scaffold** in `services/plugin/credential.py` — replaces the per-router `_SPECIAL_PROVIDER_VALIDATORS` dict; one `_LLMApiKey._probe` method serves all 10 cloud LLM providers, dedicated `_probe` overrides handle Maps + Apify, `_LocalLLM.validate` overrides for Ollama / LM Studio's 2-storage edge case. **`tests/test_plugin_self_containment.py` locks the contract** with 10 invariant classes (forbidden-imports / no-router-outside-nodes / per-plugin self-registration / registry-API sanity / stale-paths-absent / main.py-does-not-mount / WS_HANDLERS-non-empty / plugin-folder-has-node-file / typed-event-factories / package-imports-cleanly); same `inspect.getsource` introspection style as `test_credential_broadcasts.py`. Wire format unchanged — frontend WS message-type strings (`whatsapp_status`, `twitter_oauth_login`, `google_oauth_status`, `android_relay_connect`, etc.) and HTTP route paths (`/api/twitter/callback`, `/api/google/callback`, `/api/android/*`) are byte-identical post-migration.- **Typed CloudEvents factory pattern for new broadcasts** (b6aecd3): every new server→FE lifecycle broadcast follows the recipe locked by `broadcast_credential_event`, `broadcast_agent_progress`, and `broadcast_agent_capability`. (1) Add a typed factory — convention: `source = "opencompany://services/<area>"`, reverse-DNS `type = "com.opencompany.<area>.<event>"`, and `subject = <exact primary entity id>`. Put snake_case workflow/execution scope inside `data`; CloudEvents extension attribute names MUST be lowercase-alphanumeric, so the historical top-level `workflow_id`/`trigger_node_id`/`correlation_id` fields are compatibility debt and are forbidden on new contracts. (2) Add a broadcaster wrapper that emits `broadcast({type: "<wire_key>", data: event.model_dump(mode="json", exclude_none=True)})`. (3) FE validates `specversion`, ID/source/type, subject ownership, and event-specific payload invariants, deduplicates by `(source,id)`, then writes only the exact scoped entity. (4) Add contract tests in `test_events.py` and keep `tests/test_status_broadcasts.py`'s raw-callsite allowlist empty. Raw `node_status` remains allowed only as the established high-frequency latest-state/reconnect projection paired with a typed lifecycle occurrence; it is not the event of record.@@ -2778,7 +2778,7 @@ This function:- **`company clean` preserves `.opencompany/{workflows,deploy,packages}/`**: `cli/commands/clean.py` iterates the canonical `<repo>/.opencompany/` and pre-rebrand `<repo>/.machina/` roots and skips anything in `_OPENCOMPANY_KEEP = frozenset({"workflows", "deploy", "packages"})`. Wipes `claude/`, `workspaces/`, `*.db` as before. `workflows/` holds the shipped seed JSONs (git-tracked); `deploy/` holds Terraform state for LIVE cloud resources (only `company deploy destroy` removes it); `packages/` holds the OpenCompany-managed binaries (Temporal CLI ~114 MB, Stripe CLI, the shared bun packages tree) — re-fetchable but expensive, so clean+build cycles stay offline-safe cache hits. Test `cli/tests/test_clean.py::test_opencompany_keep_preserves_workflows_deploy_and_packages` locks the keep-list.- **No raw `print()` outside three sanctioned helpers**: `main._startup_log` (pre-logger boot markers), `core.container._clog` (DI-bootstrap markers), and `nodes.code.python_executor.captured_print` (the sandbox builtin handed to user code). Everything else goes through `logger = get_logger(__name__)`. The supervisor prefixes every aggregated line with `[HH:MM:SS.fff]` so no inner `TimeStamper` is needed; structlog console mode is deliberately timestamp-less. Test `server/tests/test_no_raw_prints.py` AST-walks the tree and flags any unsanctioned `print(...)` call.- **`configure_logging(settings)` must run BEFORE plugin self-registration imports**: in `main.py`, `Settings()` + `configure_logging(settings)` + `init_tracing()` + `get_logger(__name__)` happen ahead of `from core.container import container` and `from routers import …`. Otherwise plugin folders that register on import call `logger.debug(...)` while structlog is still on its default processor chain (which includes `TimeStamper` + no `filter_by_level`) — symptom is double timestamps + debug records leaking despite `LOG_LEVEL=INFO`.-- **Plugin-identifier shape validator (`services/plugin/identifiers.py`)** (`7700f87`): `NODE_TYPE_PATTERN = r"^[A-Za-z_][A-Za-z0-9_]*$"` + `is_valid_node_type(value)`. Single source of truth used by FastAPI URL routes (`Path(pattern=NODE_TYPE_PATTERN)` in `routers/schemas.py`) AND internal helpers (`nodes/_visuals.py::get_plugin_meta` / `get_plugin_icon_path`). Hardens the icon-resolution endpoints: the registry lookup `get_node_class(node_type)` already gates the taint in practice, but CodeQL can't follow registry-mediated sanitization, so `fullmatch` at the function boundary states the constraint explicitly. **It does NOT close the CodeQL `py/path-injection` alerts, despite what this line used to claim.** Verified against the live SARIF (2026-07-26): `_visuals.py:217`, `_visuals.py:239` and `schemas.py:144` are still reported on every scan and were closed by *manual dismissal*, not by the regex. The reason is structural and worth knowing before you attempt a fix: the flagged sink is the path-construction/`resolve()` call itself, so no guard placed after it can clear the taint, and an interprocedural `fullmatch` in a helper is not treated as a barrier either. **`py/path-injection` on a containment helper is dismissed here, not coded around** — see `nodes/filesystem/_backend.py::resolve_within` and alerts #29-33, #39, #140, #142, #154-157. Contract test (`tests/services/test_identifiers.py`, 47 cases) locks the regex against `../etc/passwd`, `..\windows\system32`, `foo\x00bar`, `%2e%2e%2fetc`, `foo;bar`, `${HOME}`, `` foo`bar ``, CRLF, and non-string input.+- **Plugin-identifier shape validator (`services/plugin/identifiers.py`)** (`7700f87`): `NODE_TYPE_PATTERN = r"^[A-Za-z_][A-Za-z0-9_]*$"` + `is_valid_node_type(value)`. Single source of truth used by FastAPI URL routes (`Path(pattern=NODE_TYPE_PATTERN)` in `routers/schemas.py`) AND internal helpers (`nodes/_visuals.py::get_plugin_meta` / `get_plugin_icon_path`). Hardens the icon-resolution endpoints: the registry lookup `get_node_class(node_type)` already gates the taint in practice, but CodeQL can't follow registry-mediated sanitization, so `fullmatch` at the function boundary states the constraint explicitly. **It does NOT close the CodeQL `py/path-injection` alerts, despite what this line used to claim.** Verified against the live SARIF (2026-07-26): `_visuals.py:217`, `_visuals.py:239` and `schemas.py:144` are still reported on every scan and were closed by *manual dismissal*, not by the regex. The reason is structural and worth knowing before you attempt a fix: the flagged sink is the path-construction/`resolve()` call itself, so no guard placed after it can clear the taint, and an interprocedural `fullmatch` in a helper is not treated as a barrier either. **`py/path-injection` on a containment helper is dismissed here, not coded around** — see `nodes/filesystem/_backend.py::resolve_within` and alerts #29-33, #39, #140, #142, #154-157. **The one shape the query does accept** is normalise-then-prefix-check performed *before* the path is used: `core/approot.py::resolve_static_asset` joins the request path onto the real base dir, runs `os.path.normpath`, and returns `None` unless the result starts with that base plus a separator. `main.py`'s SPA fallback was rewritten onto it in September 2026 (alerts #158-160) because that route can take a `None` and fall through to the SPA shell. Reach for that shape when the call site tolerates a miss; keep the dismissal route for helpers whose flagged sink IS the `resolve()`. Contract test (`tests/services/test_identifiers.py`, 47 cases) locks the regex against `../etc/passwd`, `..\windows\system32`, `foo\x00bar`, `%2e%2e%2fetc`, `foo;bar`, `${HOME}`, `` foo`bar ``, CRLF, and non-string input.- **CLI recovery-resilience invariants** (May 2026): `python -m cli clean` must run end-to-end on a system Python with no third-party deps. Achieved by: (a) every verb in [`cli/cli.py`](./cli/cli.py) is a lazy stub that imports its impl inside the function body — `import cli.cli` does NOT pull in `anyio` / `psutil` / `cli.supervisor` / sibling verb modules (300 → 131 modules at boot). (b) [`cli/_common.py`](./cli/_common.py) defers `cli.supervisor` imports inside `build_backend_spec` so importing `cli._common` (and therefore `clean.py`) doesn't drag in rich. (c) [`cli/platform_.py`](./cli/platform_.py) lazy-imports `platformdirs` inside the four `user_*_dir` helpers — module loads without the wheel. (d) [`cli/commands/clean.py`](./cli/commands/clean.py) uses stdlib `print()` (no rich) + lazy `cli.ports` import inside `_kill_running_processes` wrapped in `try/except ImportError` (skip-with-warning if psutil is missing). (e)+(f) retired July 2026: the CLI no longer supervises Temporal at all — the backend lifespan owns the dev server via `services.temporal._runtime.ensure_started()` (`cli/commands/_temporal_specs.py` and the `_supervised_runtime.py` shim were deleted; long-running specs spawn the server venv interpreter directly via `cli.platform_.server_venv_python`, no resident `uv run` parents). (g) `daemon` is now a verb-per-file package ([`cli/commands/daemon/`](./cli/commands/daemon/) — `_state.py` + `start.py` + `stop.py` + `status.py` + `restart.py`) following pdm's `commands/venv/` shape; PID-file resolution is a function (`pid_dir()`) not a module-level attribute so platformdirs only loads at call time.- **Wave 13 canary fixes** — six load-bearing corrections to the Wave 12 event framework. See [docs-internal/event_framework.md → Wave 13 fixes](./docs-internal/event_framework.md#wave-13-fixes) for the full breakdown. Key invariants to remember when working on canary triggers: (1) `register_canary_trigger_type(node_type, cloudevent_type)` requires the CloudEvents reverse-DNS string as second arg — must match the producer's `WorkflowEvent.type` exactly or the deployment manager's `EventType` SA won't match `dispatch.emit`'s Visibility query (silent firing failure). Diverging re-registration raises `ValueError` so plugin upgrades surface loudly. (2) Plugin `_events.py` for canary-registered triggers is canary-only — no `event_waiter.dispatch` / `send_custom_event` calls (those have zero consumers in canary-on mode). `dispatch.emit(envelope, wire_routing_key=...)` handles BOTH Temporal Signal fan-out AND in-process WS broadcast. (3) `TriggerListenerWorkflow` + `PollingTriggerWorkflow` call `broadcast_trigger_status_activity` before/after each child spawn for firing-pulse UX (matches legacy `triggers.py` collector/processor). (4) `PollingTriggerNode.as_poll_activity` returns `seen_ids: list(current)` — NOT `list(prior_seen | current)` (the latter grows unboundedly; Gmail at ~100/day hit ~36K entries in a year). The legacy `_build_poll_coroutine` does `seen = set(current)` at end of cycle for the same reason. Visibility-filtered providers (Gmail-unread) re-emit on re-surface, which is correct. (5) Canary trigger output persistence: `MachinaWorkflow.run`'s pre-executed loop schedules `store_node_output_activity` for every firing trigger so `ParameterResolver` can resolve `{{triggerNode.field}}` in downstream nodes (the legacy `_execute_from_trigger` did this via `_store_output(trigger_node_id, "output_0", ...)`; canary skipped it pre-fix). Skips non-firing siblings (`_trigger_output={"not_triggered": True}`). (6) `DeploymentManager.cancel` sweeps stuck node statuses via `_clear_stuck_node_statuses(workflow_id, include_waiting=True)` + emits terminal `update_workflow_status(executing=False, workflow_id=...)`. Without these the FE leaves downstream nodes glowing forever after deployment cancel and the toolbar Start/Stop indicator stays at `executing=True`. The delegation guard inside `_clear_stuck_node_statuses` still protects in-flight fire-and-forget child agents.- **Deployment reconcile snapshot on WS connect** (de8df87): a stale FE `deploymentStatus.isRunning=true` used to survive a backend restart because the in-memory `DeploymentManager._deployments` dict was wiped (no recovery wiring — the `RecoverySweeper.set_recovery_callback` extension point exists but is intentionally unwired; see the "workflow status not properly recognized" investigation). The Start button therefore stayed showing "Stop" forever. Fix: new `WorkflowEvent.deployment_snapshot(running_workflow_ids)` typed factory (`source = opencompany://services/workflow`, `type = workflow.deployment.snapshot`) + `broadcaster._send_deployment_snapshot(websocket)` that runs at the tail of `broadcaster.connect()` (single-target, NOT fan-out — only the just-connecting client needs to reconcile). FE `case 'deployment_snapshot'` in `WebSocketContext.tsx` iterates `useAppStore.workflowUIStates` and clears `isExecuting=true` on any workflow NOT in the snapshot's running set (empty list is meaningful — this is the load-bearing reset). Also reconciles `deploymentStatus.isRunning` for the active workflow. Distinct from `workflow_lifecycle("deployment.started")` (state-transition edge event) — the snapshot is an idempotent state dump tied to client connect, not a transition.diff --git a/docs-internal/ci_cd.md b/docs-internal/ci_cd.mdindex 77ea90a2..308dd480 100644--- a/docs-internal/ci_cd.md+++ b/docs-internal/ci_cd.md@@ -2,7 +2,7 @@Internal reference for OpenCompany's GitHub Actions setup. Workflow inventory, release flow, and the composite setup action.-The repo currently ships **three workflows** plus one composite action. A number of hardening / security workflows described in earlier revisions of this doc (CodeQL, zizmor, rollback, reusable PyPI publish, cross-platform test-install) are **not yet implemented** — see [Planned (not yet implemented)](#planned-not-yet-implemented) at the end.+The repo currently ships **five workflows** plus one composite action. Code scanning is live but has no workflow file: it runs through GitHub's **default CodeQL setup**, which is why alerts land in the Security tab with nothing in `.github/workflows/` to point at (see [Code scanning](#code-scanning)). The other hardening workflows described in earlier revisions of this doc (zizmor, rollback, reusable PyPI publish, cross-platform test-install) are **not yet implemented** — see [Planned (not yet implemented)](#planned-not-yet-implemented) at the end.---@@ -207,6 +207,41 @@ folder — both the folder and the workflow are gone.---+## Code scanning++CodeQL runs through GitHub's **default setup** (Python + JS/TS), configured+in repository settings rather than by a workflow file. There is no+`codeql.yml` to read, so the only places the configuration is visible are+the Security tab and the `CodeQL` check on each push.++Triage has two outcomes, and picking the right one matters because the+wrong one leaves a permanent lie in the code:++- **Fix it** when the call site can be restructured into a shape the query+ recognises. For `py/path-injection` that shape is normalise-then-prefix-+ check *before* the path is used: join, `os.path.normpath`, then refuse+ anything that does not start with the real base directory plus a+ separator. `core/approot.py::resolve_static_asset` is the worked example+ (alerts #158-160, September 2026).+- **Dismiss it** when the flagged sink is the containment helper itself.+ A guard placed after the `resolve()` cannot clear the taint and an+ interprocedural `fullmatch` in a helper is not treated as a barrier, so+ the alert will return on every scan no matter what is written. Dismiss+ with a comment naming the containment invariant; do not add a decorative+ check that implies the alert was addressed. `nodes/filesystem/_backend.py+ ::resolve_within` and `nodes/_visuals.py` are the standing examples.++Two findings are dismissed as by-design rather than fixed:++| Query | Where | Why |+|---|---|---|+| `js/code-injection` | `server/nodejs/src/index.ts` `/execute` | That route **is** the JavaScript executor behind the `javascriptExecutor` / `typescriptExecutor` nodes, so evaluating request-supplied code is its purpose. It binds to localhost and is spawned and called only by the same-machine backend; `vm` is documented as not a security boundary. |+| `py/path-injection` | containment helpers | See above. |++Inline `// codeql[...]` / `# codeql[...]` comments do **not** suppress+anything under default setup — several sat in the sidecar for months while+the alerts stayed open. Dismiss in the Security tab instead.+## Dependency update policy (Dependabot disabled)Dependabot opens **no pull requests**, neither version bumps nor security@@ -251,6 +286,5 @@ The following existed in earlier drafts of this doc but are **not present in the- **`publish-pypi.yml`** — reusable PyPI publish (OIDC trusted publishing, `uv build --no-sources`, `pypa/gh-action-pypi-publish`). No PyPI distribution is published today.- **`test-install.yml`** — cross-platform end-user install smoke (`bun add -g`, git clone, install script) across 3 OS.- **`rollback.yml`** — manual registry deprecate (`npm deprecate` has no bun equivalent, so this would be the one place the npm CLI reappears) + optional revert PR.-- **`codeql.yml`** — Python + JS/TS SAST (`security-extended`).- **`check-zizmor.yml`** — workflow-security linter (SARIF to the Security tab).- **`.pre-commit-config.yaml`** — ruff / prettier / eslint / actionlint hooks (note: the project rule is to verify with pytest + the root `typecheck` gate + eslint, not ruff). This file is **not currently present in the tree**; the entry describes intent, not a live hook.diff --git a/docs-internal/design-system/readme.md b/docs-internal/design-system/readme.mdindex fbe8fc9d..478f6ce4 100644--- a/docs-internal/design-system/readme.md+++ b/docs-internal/design-system/readme.md@@ -23,7 +23,7 @@ The app ships 12 visual themes (light, dark, Renaissance, Cyber, Edo, Steampunk- **Short declarative fragments for value props.** "No code required. No subscription. No usage limits."- **Sentence case everywhere** — headings ("What You Can Build" is Title Case in README headers, but UI copy is sentence case: "No workflows yet", "Create your first workflow to get started"). Buttons are single words or short verb phrases: **Start**, **Stop**, **Save**, **New Workflow**, **Apply All**.- **UI microcopy is terse and systemy.** Status bar reads like a shell prompt: `ONLINE | WF: Test | NODES: 12 | THEME: DARK | 14:02:33`. States: "Saved" / "Modified". Tooltips are full sentences: "Override all agent nodes in this workflow to use the selected model".-- **Numbers as proof.** "50+ other services", "12 LLM providers", "17 specialized agent types", "16 device services". Counts are surfaced in the UI too (badge with component count, "3 saved").+- **Numbers as proof.** "50+ other services", "13 LLM providers", "17 specialized agent types", "16 device services". Counts are surfaced in the UI too (badge with component count, "3 saved").- **Emoji: not part of the brand voice.** Marketing copy uses none. In-product the only emoji are *functional glyphs* (⚙️ gear button, 📦 fallback icon) — never decorative, never in sentences.- **Bold for product nouns** mid-sentence: "**AI Employee**", "**Stripe**", "**WhatsApp**".- Example headline pairs: "Personal AI assistants that remember" / "Task automations that run themselves" — lowercase-after-first-word, noun phrase + verb clause.diff --git a/docs-internal/desktop_app.md b/docs-internal/desktop_app.mdindex e4514244..253bc708 100644--- a/docs-internal/desktop_app.md+++ b/docs-internal/desktop_app.md@@ -40,7 +40,7 @@ OpenCompany.app / OpenCompany/ / AppImagepackage.json package.json, .opencompany/workflows/,.opencompany/workflows/ server/uv.lock, server/pyproject.tomlclient/dist/- server/ (+ nodejs/dist/index.js — the sidecar's own `bun build --target=bun`, express inlined)+ server/ (+ nodejs/dist/index.js — the sidecar's own `bun build --target=bun`, dependencies inlined)runtime/uv/uv[.exe] pinned uv (runtimes.json)python/ python-build-standalone install_only_strippeddiff --git a/docs-internal/desktop_host_contract.md b/docs-internal/desktop_host_contract.mdindex dc39f128..4d5c5026 100644--- a/docs-internal/desktop_host_contract.md+++ b/docs-internal/desktop_host_contract.md@@ -49,6 +49,14 @@ the rule: no other module under `server/` may climb above `server/` with`parents[N]` or a `.parent` chain, or spell `client/dist`, `.env.template`,`package.json` in a path join.+`resolve_static_asset(base_dir, relative)` lives here too, for the one route+that maps an attacker-controlled path onto the tree: the SPA fallback in+`main.py`. It joins, `os.path.normpath`s, and returns `None` unless the+result sits under the real base directory — so `..` traversal, an absolute+path, and a prefix sibling such as `client/dist2` all fall through to the+SPA shell instead of serving a file. Keeping it beside `client_dist()` is+what lets the route itself build no paths at all.+## 2. Env layering without the CLI`core.env_defaults.apply_file_defaults_to_environ()` runs at the top ofdiff --git a/docs-internal/node-logic-flows/ai_agents/aiAgent.md b/docs-internal/node-logic-flows/ai_agents/aiAgent.mdindex ca8207be..8b3567ba 100644--- a/docs-internal/node-logic-flows/ai_agents/aiAgent.md+++ b/docs-internal/node-logic-flows/ai_agents/aiAgent.md@@ -134,7 +134,7 @@ flowchart TDand potentially `compaction_starting` / `compaction_completed` events.`BaseNode.execute()` additionally wraps the body in a `node.aiAgent.execute`OpenTelemetry span + `log_context(node_id, node_type, workflow_id)`.-- **External API calls**: all 12 providers run through `ChatUnifier` and the+- **External API calls**: all 13 providers run through `ChatUnifier` and thenative provider layer: Anthropic uses `anthropic`, Gemini uses`google-genai`, and OpenAI plus the OpenAI-compatible providers (OpenRouter,Groq, Cerebras, xAI, DeepSeek, Kimi, Mistral, Ollama, and LM Studio) usediff --git a/docs-internal/node-logic-flows/ai_chat_models/cerebrasChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/cerebrasChatModel.mdindex c9e3c1fb..3acd8f09 100644--- a/docs-internal/node-logic-flows/ai_chat_models/cerebrasChatModel.md+++ b/docs-internal/node-logic-flows/ai_chat_models/cerebrasChatModel.md@@ -11,11 +11,12 @@## Purpose-Ultra-fast inference on Cerebras' custom AI hardware. The current curated-models are GPT-OSS-120b plus preview Z.ai GLM 4.7 and Gemma 4 tiers. The+Ultra-fast inference on Cerebras' custom AI hardware. The public endpoints+serve exactly two models: `gpt-oss-120b` (default) and `qwen-3.8-27b`; other+families exist only on Dedicated Endpoints. The`ChatModelBase.chat` operation calls`AIService.execute_chat`, which routes through `ChatUnifier`. Like Groq,-Cerebras is one of the eight OpenAI-compatible providers registered in+Cerebras is one of the nine OpenAI-compatible providers registered in`providers/_compat.py`; bare chat and current agent executions share thisnative provider path.@@ -31,12 +32,12 @@ native provider path.|------|------|---------|----------|---------------------|-------------|| `prompt` | string | `""` | yes | - | User message || `system_prompt` | string | `""` | no | - | System prompt |-| `model` | string | `""` (injected) | no | - | e.g. `llama3.1-8b`, `gpt-oss-120b`, `qwen-3-235b-a22b` |+| `model` | string | `""` (injected) | no | - | `gpt-oss-120b` (default) or `qwen-3.8-27b` || `temperature` | number\|null | `null` | no | - | Narrower range than OpenAI (0-1.5 rather than 0-2) |-| `max_tokens` | number\|null | `null` (up to 8K) | no | - | 1-200000 |+| `max_tokens` | number\|null | `null` (up to 40K) | no | - | 1-200000 || `top_p` | number\|null | `1.0` | no | - | |-| `thinking_enabled` | boolean | `false` | no | - | Only Qwen-3-235b supports format-based reasoning |-| `thinking_budget` | number\|null | `2048` | no | `thinking_enabled=[true]` | 1024-16000 (Cerebras Qwen budget) |+| `thinking_enabled` | boolean | `false` | no | - | Inert: no current Cerebras model is in `thinking_models` |+| `thinking_budget` | number\|null | `2048` | no | `thinking_enabled=[true]` | 1024-16000; inert while `thinking_models` is empty || `reasoning_format` | enum | `parsed` | no | - | `parsed` / `hidden` - same semantics as Groq Qwen (inherited base field) || `api_key` | string\|null | `null` (injected) | no | - | `auth_service.get_api_key('cerebras', 'default')` |diff --git a/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.mdindex 9ee1f091..bc670fd4 100644--- a/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md+++ b/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md@@ -11,7 +11,7 @@## Purpose-DeepSeek V4.1 models (`deepseek-flash`, `deepseek-v4.1-flash`, `deepseek-v4-pro`); `deepseek-v4-flash` is a retired alias still served by V4.1-Flash, and `deepseek-chat` / `deepseek-reasoner` were discontinued in July 2026. Uses the OpenAI-compatible DeepSeek endpoint via the `services/llm/providers` layer (native path). The `ChatModelBase.chat` operation calls `AIService.execute_chat`. The plugin docstring describes `deepseek-chat`/`deepseek-reasoner` (V3) but the registry description in `__init__.py` predates the V4 rename - the card's V4 listing reflects current `llm_defaults.json`.+DeepSeek V4.1 models (`deepseek-flash`, `deepseek-v4.1-flash`, `deepseek-v4-pro`); `deepseek-v4-flash` is a retired alias still served by V4.1-Flash, and `deepseek-chat` / `deepseek-reasoner` were discontinued in July 2026. Uses the OpenAI-compatible DeepSeek endpoint via the `services/llm/providers` layer (native path). The `ChatModelBase.chat` operation calls `AIService.execute_chat`. The node's registry description was corrected to the V4.1 line in September 2026; `llm_defaults.json` stays the source of truth for the curated ids.## Inputs (handles)@@ -46,7 +46,7 @@ DeepSeek V4.1 models (`deepseek-flash`, `deepseek-v4.1-flash`, `deepseek-v4-pro````ts{response: string;- thinking: string | null; // reasoning_content from deepseek-reasoner; null for deepseek-chat+ thinking: string | null; // reasoning_content when the model emits one; null otherwisethinking_enabled: boolean;model: string;provider: 'deepseek';@@ -78,7 +78,7 @@ flowchart TD- **Provider routing**: `detect_ai_provider` matches `'deepseek' in node_type.lower()` **first** (before kimi/mistral/cerebras/groq/openrouter/anthropic/gemini), so routing is unambiguous.- **Native path**: uses the OpenAI SDK with DeepSeek's base URL from `llm_defaults.json`. The compatible provider sends `max_tokens` directly in the Chat Completions request; there is no LangChain parameter translation on current chat or agent executions.- **Model ID handling**: only the UI-only `[FREE] ` decoration is stripped. The remaining model ID is preserved.-- **`deepseek-reasoner` always-on CoT**: reasoning_content is ALWAYS produced, regardless of `thinkingEnabled`. The native provider extracts it into `LLMResponse.thinking`.+- **Reasoning trace**: when DeepSeek returns `reasoning_content` the native provider extracts it into `LLMResponse.thinking`, regardless of `thinkingEnabled`. (The always-on-CoT `deepseek-reasoner` this rule was written for was discontinued in July 2026; the extraction still applies to any model that emits the field.)## Side Effects@@ -97,9 +97,9 @@ flowchart TD## Edge cases & known limits-- **`deepseek-reasoner` always thinks**: `thinkingEnabled=false` does NOT disable the reasoning trace; it just means the UI won't highlight it. The response still contains `reasoning_content`.+- **`thinkingEnabled=false` does not suppress a trace**: it only means the UI will not highlight one. If the model returns `reasoning_content`, the response still carries it.- **`thinkingBudget` has no effect**: DeepSeek reasoning is not budget-configurable; the field is silently ignored.-- **Context and output**: the curated V4 models use a 1,048,576-token context window and a 65,536-token output ceiling; legacy `deepseek-chat` and `deepseek-reasoner` remain 131,072-context aliases.+- **Context and output**: the curated models use a 1,048,576-token context window and the documented 384,000-token output ceiling. `deepseek-v4-flash` is a retired name DeepSeek still accepts and serves with V4.1-Flash at the Flash price.- **OpenAI-compatible but not OpenAI**: features like `response_format: json_object` have subtly different behavior.- **Error boundary**: typed OpenAI SDK failures become user-safe `NodeUserError` values in `ChatUnifier` and are re-raised to `BaseNode.execute()`, which produces the standard failure envelope. Unexpected failures are logged and returned by `execute_chat`.diff --git a/docs-internal/node-logic-flows/ai_chat_models/groqChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/groqChatModel.mdindex 17b39681..6fe06a51 100644--- a/docs-internal/node-logic-flows/ai_chat_models/groqChatModel.md+++ b/docs-internal/node-logic-flows/ai_chat_models/groqChatModel.md@@ -14,7 +14,7 @@Ultra-fast inference via Groq's LPU hardware. Models include Llama 3.x / 4,Qwen3-32b, GPT-OSS. The `ChatModelBase.chat` operation calls`AIService.execute_chat`, which routes through `ChatUnifier`; Groq is one of-the eight OpenAI-compatible providers registered in `providers/_compat.py`.+the nine OpenAI-compatible providers registered in `providers/_compat.py`.Bare chat and current agent executions share this native provider path.## Inputs (handles)@@ -29,11 +29,11 @@ Bare chat and current agent executions share this native provider path.|------|------|---------|----------|---------------------|-------------|| `prompt` | string | `""` | yes | - | User message || `system_prompt` | string | `""` | no | - | System prompt |-| `model` | string | `""` (injected) | no | - | e.g. `llama-3.1-70b-versatile`, `qwen/qwen3.8-27b`, `groq/compound-beta` |+| `model` | string | `""` (injected) | no | - | e.g. `openai/gpt-oss-120b` (default), `llama-3.3-70b-versatile`, `qwen/qwen3.8-27b` || `temperature` | number\|null | `null` | no | - | 0-2 || `max_tokens` | number\|null | `null` (8-131K per model) | no | - | 1-200000 || `top_p` | number\|null | `1.0` | no | - | |-| `thinking_enabled` | boolean | `false` | no | - | Only Qwen3-32b supports reasoning |+| `thinking_enabled` | boolean | `false` | no | - | Only the Qwen3 family supports format-based reasoning; GPT-OSS takes `reasoning_effort` || `reasoning_format` | enum | `parsed` | no | `thinking_enabled=[true]` | `parsed` (returns reasoning) or `hidden` (suppresses it) || `api_key` | string\|null | `null` (injected) | no | - | `auth_service.get_api_key('groq', 'default')` |@@ -84,7 +84,7 @@ flowchart TD- **Native OpenAI-compatible path**: `ChatUnifier` resolves the `groq` specregistered in `providers/_compat.py` (reuses `OpenAIProvider` with the`base_url` from `llm_defaults.json`) for both chat and agent requests.-- **Reasoning**: only Qwen3-32b actually honors `reasoningFormat`. Non-Qwen models ignore the flag.+- **Reasoning**: only the Qwen3 family honors `reasoningFormat` (`qwen/qwen3.8-27b` today; Groq retired `qwen/qwen3-32b`). Other models ignore the flag, and GPT-OSS takes `reasoning_effort` instead.- **Model ID handling**: only the UI-only `[FREE] ` decoration is stripped. Owner-qualified IDs such as `openai/gpt-oss-120b` and `qwen/qwen3.8-27b` are preserved because Groq requires them.## Side Effectsdiff --git a/docs-internal/node-logic-flows/ai_chat_models/kimiChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/kimiChatModel.mdindex fd46a388..b0957acb 100644--- a/docs-internal/node-logic-flows/ai_chat_models/kimiChatModel.md+++ b/docs-internal/node-logic-flows/ai_chat_models/kimiChatModel.md@@ -13,8 +13,8 @@Kimi models by Moonshot AI. `kimi-k3` is the current default with a1,048,576-token context window and a 131,072-token output ceiling;-`kimi-k2.6`, `kimi-k2.5`, and `kimi-k2.7-code` remain available as 262K-tiers. The node uses Moonshot's OpenAI-compatible endpoint through the native+`kimi-k2.6`, `kimi-k2.7-code` and `kimi-k2.7-code-highspeed` remain available+as 262K tiers (`kimi-k2.5` left Moonshot's pricing page in September 2026). The node uses Moonshot's OpenAI-compatible endpoint through the nativeprovider layer and the shared `ChatModelParams`.## Inputs (handles)@@ -29,11 +29,11 @@ provider layer and the shared `ChatModelParams`.|------|------|---------|----------|---------------------|-------------|| `prompt` | string | `""` | yes | - | User message || `system_prompt` | string | `""` | no | - | System prompt |-| `model` | string | `""` (injected) | no | - | `kimi-k3` default; `kimi-k2.6`, `kimi-k2.5`, and `kimi-k2.7-code` also supported |-| `temperature` | number\|null | `null` | no | - | K2.5/K2.6/K2.7 Code force 0.6; K3 uses the supplied/default value clamped to 0-1 |+| `model` | string | `""` (injected) | no | - | `kimi-k3` default; `kimi-k2.6`, `kimi-k2.7-code` and `kimi-k2.7-code-highspeed` also supported |+| `temperature` | number\|null | `null` | no | - | K2.6 / K2.7 Code (+ `-highspeed`) force 0.6; K3 uses the supplied/default value clamped to 0-1 || `max_tokens` | number\|null | `null` (model ceiling) | no | - | Up to 131,072 for K3; lower model-specific ceilings for K2 tiers || `top_p` | number\|null | `1.0` | no | - | |-| `thinking_enabled` | boolean | `false` (Params default) | no | - | When false/unset, the native provider explicitly disables K2.5/K2.6/K2.7 thinking defaults |+| `thinking_enabled` | boolean | `false` (Params default) | no | - | When false/unset, the native provider explicitly disables the K2 tiers' thinking defaults || `api_key` | string\|null | `null` (injected) | no | - | `auth_service.get_api_key('kimi', 'default')` |(Kimi uses the shared `ChatModelParams` unchanged; field names are snake_case, unknown keys ignored.)diff --git a/docs-internal/plugin_system.md b/docs-internal/plugin_system.mdindex 77d5224b..46e31f08 100644--- a/docs-internal/plugin_system.md+++ b/docs-internal/plugin_system.md@@ -846,8 +846,8 @@ All Wave 10 invariants in `test_node_spec.py` still run; Wave 11 invariants in `TelegramCredential + ApifyCredential + 12 LLM providers + 3 inlinesearch credentials + Stripe / Vercel / GitHub / Cloudflare /WhatsApp); 29 plugins declared `credentials = (...)`. Today the live- numbers are `len(CREDENTIAL_REGISTRY)` (34) and 53 plugin files with a- non-empty `credentials` tuple. Agents stay poly-provider (empty tuple).+ numbers are `len(CREDENTIAL_REGISTRY)` (35) and 55 node types with a+ non-empty `credentials` tuple (September 2026). Agents stay poly-provider (empty tuple).- Wave 11.E.1 — Modularised credentials into per-domain`nodes/<group>/_credentials.py` files. `server/credentials/`directory deleted; auto-discovery rides on node-package import.diff --git a/docs-internal/pricing_service.md b/docs-internal/pricing_service.mdindex cbc6a734..b36a7c40 100644--- a/docs-internal/pricing_service.md+++ b/docs-internal/pricing_service.md@@ -67,17 +67,17 @@ Located at `server/config/pricing.json`. User-editable; picked up via `PricingSe```json{- "version": "2026.07",- "last_updated": "2026-07-25",+ "version": "2026.09",+ "last_updated": "2026-09-15","llm": {"openai": {- "gpt-5": {"input": 1.25, "output": 10.00},- "o3": {"input": 2.00, "output": 8.00, "reasoning": 8.00},- "_default": {"input": 2.50, "output": 10.00}+ "gpt-5.6-sol": {"input": 4.00, "output": 20.00, "cache_read": 0.40},+ "gpt-5.6-luna": {"input": 0.20, "output": 1.20, "cache_read": 0.02},+ "_default": {"input": 2.50, "output": 15.00}},"anthropic": {- "claude-opus-4.6": {"input": 5.00, "output": 25.00, "cache_read": 0.50},+ "claude-opus-5": {"input": 5.00, "output": 25.00, "cache_read": 0.50},"_default": {"input": 3.00, "output": 15.00, "cache_read": 0.30}}},@@ -166,23 +166,23 @@ from services.pricing import get_pricing_servicepricing = get_pricing_service()# Get pricing for a model (partial matching supported)-model_pricing = pricing.get_pricing('anthropic', 'claude-3-5-sonnet-20241022')-# Returns ModelPricing(input_per_mtok=3.00, output_per_mtok=15.00, ...)+model_pricing = pricing.get_pricing('anthropic', 'claude-sonnet-5')+# Returns ModelPricing(input_per_mtok=2.00, output_per_mtok=10.00, ...)# Calculate cost for token usagecost = pricing.calculate_cost(provider='anthropic',- model='claude-3-5-sonnet',+ model='claude-sonnet-5',input_tokens=5000,output_tokens=1500,cache_read_tokens=200)# Returns: {-# 'input_cost': 0.015,-# 'output_cost': 0.0225,-# 'cache_cost': 0.00006,+# 'input_cost': 0.01,+# 'output_cost': 0.015,+# 'cache_cost': 0.00004,# 'reasoning_cost': 0.0,-# 'total_cost': 0.03756+# 'total_cost': 0.02504# }```diff --git a/server/nodes/README.md b/server/nodes/README.mdindex e61e17c2..40621ac2 100644--- a/server/nodes/README.md+++ b/server/nodes/README.md@@ -201,7 +201,7 @@ the sibling file via relative path:from ._credentials import GoogleCredential # shared with 6 siblings# inside server/nodes/model/openai_chat_model/__init__.py-from ._credentials import OpenAICredential # one of 10 cloud LLM creds+from ._credentials import OpenAICredential # one of 11 cloud LLM creds# inside server/nodes/twitter/twitter_send/__init__.pyfrom ._credentials import TwitterCredential # shared with 3 siblings@@ -372,7 +372,7 @@ The credential-validator dispatch is a sibling concern, handled by theexisting `services/plugin/credential.py:Credential` base class. Your`Credential` subclass overrides `_probe(api_key) -> ProbeResult` (or,in rare cases like local-LLM 2-storage, the whole `validate(data)--> dict` classmethod). Maps, Apify, all 10 cloud LLM providers, and+-> dict` classmethod). Maps, Apify, all 11 cloud LLM providers, andboth local-LLM providers (Ollama / LM Studio) all dispatch through thesame scaffold — no `_SPECIAL_PROVIDER_VALIDATORS` dict in`routers/websocket.py`.diff --git a/server/nodes/model/deepseek_chat_model/__init__.py b/server/nodes/model/deepseek_chat_model/__init__.pyindex 21c006a0..e3cc04b7 100644--- a/server/nodes/model/deepseek_chat_model/__init__.py+++ b/server/nodes/model/deepseek_chat_model/__init__.py@@ -25,7 +25,7 @@ class DeepseekChatModelNode(ChatModelBase):display_name = "DeepSeek"subtitle = "Chat Model"group = ("model",)- description = "DeepSeek V3 models (deepseek-chat, deepseek-reasoner with always-on CoT)"+ description = "DeepSeek V4.1 models (deepseek-flash, deepseek-v4-pro) with 1M context"credentials = (DeepSeekCredential,)Params = DeepseekChatModelParamsdiff --git a/server/nodes/model/kimi_chat_model/__init__.py b/server/nodes/model/kimi_chat_model/__init__.pyindex 503ad210..59b1beec 100644--- a/server/nodes/model/kimi_chat_model/__init__.py+++ b/server/nodes/model/kimi_chat_model/__init__.py@@ -8,5 +8,5 @@ class KimiChatModelNode(ChatModelBase):display_name = "Kimi"subtitle = "Chat Model"group = ("model",)- description = "Kimi K2 models by Moonshot AI with 256K context (thinking on by default)"+ description = "Kimi models by Moonshot AI (K3 1M context; K2 tiers 256K, thinking on by default)"credentials = (KimiCredential,)