7.2 KiB
AGENTS.md
- Read
CONTRIBUTING.mdbefore making non-trivial changes. - For day-to-day development and feature work, follow the development-environment workflow rather than defaulting to
setup.sh/setup.ps1. - Avoid using the setup scripts during normal feature work unless the user explicitly asks for them. Users configure
.envusually. - Try to follow red/green TDD
Check existing dev prerequisites first
For feature work, do not assume the environment needs to be recreated.
- Check whether the user already has a Python virtual environment such as
venv/or.venv/. - Check whether Postgres is already running and reachable via
POSTGRES_URI(the canonical user-data store). - Check whether Redis is already running.
- Reuse what is already working. Do not stop or recreate Postgres, Redis, or the Python environment unless the task is environment setup or troubleshooting.
MongoDB is not required for the default install. It is only needed if the user opts into the Mongo vector-store backend (
VECTOR_STORE=mongodb) or is running the one-shotscripts/db/backfill.pyto migrate existing user data from the legacy Mongo-based install. In those cases,pymongois available as an optional extra, not a core dependency.
Normal local development commands
Use these commands once the dev prerequisites above are satisfied.
Backend
source .venv/bin/activate # macOS/Linux
uv pip install -r application/requirements.txt # or: pip install -r application/requirements.txt
Run the API. For local dev, prefer the ASGI entrypoint under uvicorn — it serves the whole app, matches production, and hot-reloads:
uvicorn application.asgi:asgi_app --host 0.0.0.0 --port 7091 --reload
flask --app application/app.py run --host=0.0.0.0 --port=7091 is a faster
inner loop (quick startup, the Werkzeug interactive debugger), but it serves
only the WSGI Flask app and omits the routes mounted on the ASGI shell
in application/asgi.py:
- the
/mcpFastMCP endpoint, and - the native-async SSE reconnect reader
GET /api/messages/<id>/events.
Under flask run those paths 404. Chat still works (POST /stream is a
Flask route), but a stream interrupted by a disconnect won't auto-resume on
reconnect. Use flask run only when you don't need those routes.
Production uses gunicorn -k uvicorn_worker.UvicornWorker against the same
application.asgi:asgi_app target; see application/Dockerfile for the
full flag set.
Run the Celery worker in a separate terminal (if needed):
celery -A application.app.celery worker -l INFO
On macOS, prefer the solo pool for Celery:
python -m celery -A application.app.celery worker -l INFO --pool=solo
A bare worker (no -Q) consumes every configured queue, so one worker does the
whole job — app tasks and document parsing (the read_document tool / workflow
native-file parse) alike. Use -Q only to split load: run the main worker with
-Q docsgpt and a dedicated (e.g. GPU-enabled) parser worker with -Q parsing
for heavy OCR.
Frontend
Install dependencies only when needed, then run the dev server:
cd frontend
npm install --include=dev
npm run dev
Docs site
cd docs
npm install
Python / backend changes validation
ruff check .
python -m pytest
Frontend changes
cd frontend && npm run lint
cd frontend && npm run build
Documentation changes
cd docs && npm run build
If Vale is installed locally and you edited prose, also run:
vale .
Repository map
application/: Flask backend, API routes, agent logic, retrieval, parsing, security, storage, Celery worker, and WSGI entrypoints.tests/: backend unit/integration tests and test-only Python dependencies.frontend/: Vite + React + TypeScript application.frontend/src/: main UI code, includingcomponents,conversation,hooks,locale,settings,upload, and Redux store wiring instore.ts.docs/: separate documentation site built with Next.js/Nextra.extensions/: integrations and widgets — currently the Chatwoot webhook bridge and the React widget (published to npm asdocsgpt). The Discord bot, Slack bot, and Chrome extension have been moved to their own repos underarc53/.deployment/: Docker Compose variants and Kubernetes manifests.
Coding rules
Backend
- Follow PEP 8 and keep Python line length at or under 120 characters.
- Use type hints for function arguments and return values.
- Add Google-style docstrings to new or substantially changed functions and classes.
- Add or update tests under
tests/for backend behavior changes. - Keep changes narrow in
api,auth,security,parser,retriever, andstorageareas.
Backend Abstractions
- LLM providers implement a common interface in
application/llm/(add new providers by extending the base class). - Vector stores are abstracted in
application/vectorstore/. - Parsers live in
application/parser/and handle different document formats in the ingestion stage. - Agents and tools are in
application/agents/andapplication/agents/tools/. - Celery setup/config lives in
application/celery_init.pyandapplication/celeryconfig.py. - Settings and env vars are managed via Pydantic in
application/core/settings.py.
Frontend
- Follow the existing ESLint + Prettier setup.
- Prefer small, reusable functional components and hooks.
- If shared state must be added, use Redux rather than introducing a new global state library.
- Avoid broad UI refactors unless the task explicitly asks for them.
- Do not re-create components if we already have some in the app.
Icons
DocsGPT historically mixed three icon sources: lucide-react, inline SVG components, and
.svg assets loaded via <img src=…>. For new code:
- Prefer
lucide-reactfor standard UI affordances (close, chevron, search, trash, plus, etc.). It tokenizes viacurrentColor, ships tree-shaken icons, and the codebase already imports it in 30+ places.<X className="size-4" />,<ChevronDown />, etc. - Use
assets/<name>.svg?reactwhen you need a brand-specific or domain illustration that doesn't exist in lucide (the app logo, robot fallback, retry arrow, send arrow, etc.). Always setfill="currentColor"/stroke="currentColor"in the SVG file so consumers can theme via Tailwind text classes. - Avoid
<img src={Asset}>for new icons. It blockscurrentColortheming and forces dark-variant duplicates (the audit removed several orphan dark/purple/white variants in this branch). The pattern is acceptable for existing call sites — don't bulk-migrate without a reason.
Three pre-existing dark-variant pairs (documentation, no-files, science-spark) are
hand-tuned multi-color illustrations, not pure inverts; they keep their -dark companion
files until a per-illustration refactor.
PR readiness
Before opening a PR:
- run the relevant validation commands above
- confirm backend changes still work end-to-end after ingesting sample data when applicable
- clearly summarize user-visible behavior changes
- mention any config, dependency, or deployment implications
- Ask your user to attach a screenshot or a video to it