deploy-configure
Guides DocsGPT development, deployment, service topology, model configuration, authentication, storage, observability, and upgrades.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- deploy-configure
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guides DocsGPT development, deployment, service topology, model configuration, authentication, storage, observability, and upgrades.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/auth-storage-observability.md3.4 KB
- 参考文档references/configuration.md3.9 KB
- 参考文档references/development-and-deployment.md3.5 KB
- 参考文档references/troubleshooting.md2.8 KB
- 脚本scripts/check_services.py3.1 KB
- 脚本scripts/validate_model_catalog.py6.1 KB
03 SKILL.md 正文
Deploy and Configure DocsGPT
Use this sub-skill for environment setup, application topology, provider/model configuration, auth, storage, database, queue, observability, and upgrade work.
Choose the path
- Source development: read development and deployment. Reuse an existing Python environment, Postgres, and Redis when healthy; use explicit dependency commands rather than setup bootstrap scripts.
- Container or Kubernetes deployment: read the same reference for service topology, health gates, worker queues, reverse proxy, and rollout order.
- Environment/model configuration: read configuration, then validate custom model YAML before restart.
- OIDC, SCIM, RBAC, S3, migrations, or observability: read auth, storage, and observability.
- Failure or upgrade: read troubleshooting before mutating data or identity configuration.
Baseline source-development workflow
-
Confirm Python 3.12, a project-local environment, and existing service health.
-
Configure a root
.env; do not commit it. -
Install the bundled 0.18.0 backend dependency snapshot explicitly (run from this sub-skill directory):
python -m pip install -r ../../references/backend-requirements-0.18.0.txtFor another DocsGPT release, compare repository provenance and refresh the skill rather than silently mixing dependency versions.
-
Run the full ASGI target:
uvicorn application.asgi:asgi_app --host 0.0.0.0 --port 7091 --reload -
Start a worker in another terminal when ingestion, parsing, schedules, titles, or other background jobs are needed:
celery -A application.app.celery worker -l INFOOn macOS, use
python -m celery -A application.app.celery worker -l INFO --pool=solo. -
Run the frontend separately with its local npm dependencies.
-
Verify
/api/health,/api/config, one model listing, and one bounded task before broader testing.
Flask-only development is a reduced mode: /mcp and native-async message-event reconnect are absent.
Service preflight
Use the bundled checker before blaming application code:
python scripts/check_services.py \
--postgres-uri "$POSTGRES_URI" \
--redis-url "$CELERY_BROKER_URL" \
--api-url http://localhost:7091
The helper performs connection/readiness probes only. Omit secrets from pasted output.
Model catalog changes
Use operator YAML through MODELS_CONFIG_DIR for OpenAI-compatible endpoints or extensions to registered providers. Preserve model ids after users select them.
python scripts/validate_model_catalog.py ./models-config
Validation checks shape, duplicate ids, supported capability fields, attachment aliases, reasoning effort, and OpenAI-compatible metadata. It does not contact providers.
Production rules
- Run the ASGI app through an ASGI-capable worker. Forward SSE without proxy buffering and preserve long request timeouts deliberately.
- Use Postgres as the user-data store. Set
AUTO_CREATE_DB=falseandAUTO_MIGRATE=falsewhen migrations are controlled by CI/CD. - Use Redis URLs consistently across web and workers. A worker without
-Qconsumes all configured queues; splitdocsgptandparsingonly with matching workers. - Treat
LOCAL_MODE_ADMIN=trueas local no-auth mode only; never use it on a networked deployment. - Mount model catalogs read-only, keep secrets in the platform secret store, and separate public base URLs from internal service URLs.
- Back up Postgres and object storage before schema or data migrations.
- Add optional services only for selected features: pgvector/vector database, S3-compatible storage, sandbox runner, OAuth connectors, telemetry collector.
Cross-skill routes
- Source ingestion and worker behavior: ingest-sources
- Vector-store and retrieval compatibility: retrieval-vectorstores
- Client/API verification: api-client-operations
- Sandbox and remote-device services: tools-integrations
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).