[Iamsdt]

10xgraph-api 1.0.0rc2

Second 1.0 release candidate. It is the 1.0.0rc1 code plus a round of fixes found in the pre-1.0 review and in live runs on a production stack built with 10xgraph build --docker-compose (gunicorn, JWT with scopes, ownership authorization, Postgres + Redis PgCheckpointer, Redis rate limit and token revocation, Qdrant memory). Requires 10xgraph>=1.0.0rc2,<2.0; aligned with 10xgraph-client 1.0.0-rc.2.

Full list: CHANGELOG.md section [1.0.0rc2]. Human-facing summary: RELEASE_NOTES.md.

Breaking (read before merging)

Request validation

• Ids in paths and bodies must match ^[A-Za-z0-9._:@-]{1,128}$ (422 otherwise). Snowflake ids and UUIDs fit.
• initial_state, config and resume nested deeper than 64 levels get 422.
• Uploads over the new default limits get 413: images over 50,000,000 pixels, PDFs over 10 MB or 500 pages.

Status codes and responses

• Resume errors answer 409 (wrong or stale interrupt id, thread not paused) or 422 (value fails response_schema). On /v1/graph/stream the 409 now comes before the stream starts.
• Writes during a run return 409: state put/delete, message add/delete, thread delete and /v1/graph/fix.
• DELETE /v1/threads/{id} can return success=false with data.cleanup_failed when file, trace or cache cleanup failed. The thread rows are still deleted.
• A run that hits its recursion limit answers 422 GRAPH_RECURSION_LIMIT, not 500.
• Graph validation errors use only VALIDATION_ERROR.
• POST /v1/graph/stream is labelled application/x-ndjson and sends a keep-alive line every 15 s.

Auth and authorization

• JWT_ALGORITHM must be on an allowlist, and startup fails without PyJWT when "auth": "jwt".
• auth must be exactly "jwt".
• The graph:setup scope is gone.
• AG-UI and /v1/graph/live follow runs.thread_ids like REST: under "server", an id the server never issued is refused. Frontends that make their own ids (CopilotKit) must set "runs": {"thread_ids": "client"}.

CLI

• 10xgraph build refuses --docker-compose together with --k8s.

Deploy notes

• Postgres: core schema migrations v4 and v5 run on first use. Stop old pods, let one new pod migrate, then start the rest. /health/ready fails while the schema is behind.
• Gunicorn: the generated command uses TenxGraphUvicornWorker. Regenerate Docker and Kubernetes files with 10xgraph build. Keep --graceful-timeout above the sum of the SHUTDOWN_*_SECONDS settings.
• Dependency floors: python-multipart>=0.0.31, starlette>=1.3.1, PyJWT>=2.15.0, OpenTelemetry >=1.45.1.
• Alerting: handled 4xx errors log one WARNING, not an ERROR with a traceback.

Highlights

Graceful shutdown

• The lifespan calls graph.aclose() with SHUTDOWN_GRACE_SECONDS, SHUTDOWN_STOP_SECONDS and SHUTDOWN_TIMEOUT_SECONDS.
• Cut runs are saved INTERRUPTED, and the thread resumes with the next request.
• Runs refused while closing answer 503 with Retry-After.
• HTTP streams end with a final terminal chunk on shutdown.
• Stop and shutdown now reach a REST stream after its first chunk.
• A disconnected stream frees its thread only after the run is saved.

Production setup

• 10xgraph init --template production defaults to JWT, ownership authorization and Redis rate limiting.
• 10xgraph build and 10xgraph config warn about a missing auth or authorization setup.
• The 10xgraph config editor is redesigned.

New opt-ins

• Idempotency keys.
• JWT key rotation (JWT_SECRET_KEYS) and JWT_MAX_TOKEN_AGE_SECONDS.
• 10xgraph revoke.
• REDIS_KEY_PREFIX.
• Media limits.
• Run trace.

Other additions

• Retry-After on 429 and 409.
• Typed server errors.
• A session_started event on /v1/graph/live.

Evals

• 10xgraph eval writes an HTML report and opens it, shows live progress lines, and records the pass-rate gate in the report.
• Reports carry each case's expectations.
• The init eval templates can pass, and the template weather tool answers per city.

Reliability

• Concurrency leases and revocation use the Redis clock.
• The Redis client has timeouts and a bounded pool.
• Readiness checks the schema version.
• Realtime sessions get their own state per connection.
• The default and template Live model is now gemini-3.8-live.

Security

• Message add and state put check file_id media ownership like invoke.
• Redis keys are built from percent-encoded parts, so ids cannot collide.
• Uploads are authenticated before the body is parsed.
• Client dicts and lists are size-capped.
• Secrets and request text stay out of validation errors, the access log and log lines.
• Extracted document text is deleted with its file.
• IS_DEBUG is removed.

Testing

• Unit suite: 2517 passed, 1 skipped.
• Live on the production stack (2026-10-30):
  • JWT with scopes ("Missing required scope" without them);
  • two users see only their own threads and memories;
  • Stop, docker compose stop and docker kill mid-run each leave a usable thread;
  • Redis stopped mid-run, with the run finishing and the thread working after Redis returned;
  • message and thread delete on Postgres;
  • one user in two tabs and a second browser on Gemini Live, each session with its own transcript and thread;
  • OpenAI, Gemini and Anthropic agents served through the API.
• Lint and mypy: run before merge.

[codecov]

Codecov Report

❌ Patch coverage is 87.38574% with 276 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
tenxgraph_api/src/app/utils/run_trace.py	82.26%	27 Missing and 23 partials ⚠️
tenxgraph_api/src/app/routers/graph/router.py	81.43%	38 Missing and 11 partials ⚠️
tenxgraph_api/src/app/routers/graph/idempotency.py	76.25%	32 Missing and 6 partials ⚠️
...pi/src/app/routers/graph/services/graph_service.py	90.72%	13 Missing and 1 partial ⚠️
tenxgraph_api/cli/commands/revoke.py	73.80%	10 Missing and 1 partial ⚠️
tenxgraph_api/src/app/uvicorn_worker.py	73.80%	8 Missing and 3 partials ⚠️
tenxgraph_api/src/app/core/shutdown.py	90.00%	5 Missing and 4 partials ⚠️
tenxgraph_api/cli/commands/init.py	80.48%	7 Missing and 1 partial ⚠️
tenxgraph_api/src/app/core/concurrency.py	85.10%	6 Missing and 1 partial ⚠️
...graph_api/src/app/core/exceptions/server_errors.py	91.95%	5 Missing and 2 partials ⚠️
... and 26 more

📢 Thoughts on this report? Let us know!

[Mothilal-M]

LGTM! Everything looks solid. Approving.