data-agent

FastAPI + CrewAI multi-agent data assistant.

English | 中文

What it does

Two endpoints exposed by dataagent.main:

• POST /api/analyze — runs a 3-agent CrewAI flow (researcher → analyst → writer) on a user query
• POST /api/knowledge/query — runs a single knowledge-manager agent

There is also a python -m dataagent.cli interactive shell for the same two flows.

The rest of the code (core/, routers/, app/, services/) is the supporting layer: auth, knowledge base, settings, conversations, websocket, workspace files, etc.

Run

cd dataagent
pip install -r requirements.txt
export OPENAI_API_KEY=sk-...
python main.py            # API on :8000
# or
python cli.py             # interactive shell

Docs UI: http://localhost:8000/docs Health: http://localhost:8000/health

Layout

dataagent/
├── main.py            # FastAPI entry
├── cli.py             # CLI entry
├── config.py          # pydantic-settings, env-driven
├── core/              # config, db, auth, knowledge, websocket, workspace, logging
├── routers/           # FastAPI routes
├── services/          # document service, polish service
├── agents/            # CrewAI agent factories
├── crews/             # CrewAI crew definitions
├── app/knowledge/     # lightweight KB used by routers/knowledge.py
├── config/            # example TOML/JSON configs
├── static/            # frontend assets (not currently served by main.py)
└── templates/         # placeholder index.html

Configuration

Settings come from environment variables (see dataagent/config.py) and from optional TOML files in config/. Provider examples are in config/config.example-model-*.toml.

Notes

• The frontend in static/ and templates/ is not wired up to main.py yet. API consumers should hit the JSON endpoints directly.
• Most "advanced" features in the codebase (browser automation, MCP, multi-modal extensions, agent market) are scaffold-only. The two endpoints above are what actually works end-to-end.

License

MIT