Ada

Personal agent cockpit. Runs agents (starting with a secretary), and it's where you watch them work. See ARCHITECTURE.md for the design and ROADMAP.md for the build plan.

Stack

FastAPI + Pydantic AI (agent loop) · React/TS/Vite (dashboard) · Postgres/Redis/Qdrant (Docker) · Claude (orchestrator) + Qwen-14B local via Ollama (cheap subtasks).

Run it (dev)

1. Start the stateful services (leave running):

docker compose up -d

2. Backend — in backend/:

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env        # then paste your ANTHROPIC_API_KEY
uvicorn app.main:app --reload --port 8000

3. Frontend — in frontend/:

npm install
npm run dev                 # http://localhost:5173

Open the dashboard, type a request (e.g. "add a task to prep for the Joby panel, then list my tasks"), and watch the trace panel stream Ada's steps live.

CLI trace (same event stream, terminal-side) — in backend/ with the venv active and the backend running:

python cli.py "what tasks do I have?"

Optional: local model

Install Ollama, then ollama pull qwen2.5:14b. Call local_complete() from inside a tool to route cheap work to the 5080 instead of Claude. Wired but not yet used in M1.

Where things live

• backend/app/runtime/ — the event schema + bus + supervisor (the spine)
• backend/app/agents/ — Ada agent + model router
• backend/app/tools/ — tools (Tasks now; Calendar/Gmail/etc. drop in here)
• frontend/src/App.tsx — the command-deck dashboard
• frontend/src/lib/ada.ts — WebSocket client + event types

Next

You're at Milestone 1. Next steps, in order: swap Tasks to Postgres, add the Calendar tool, add Gmail. Then M2 (multiple agents + Fleet view). See ROADMAP.md.