Real-time mind-map ideation engine. Speak (alone or in groups), and a graph of ideas builds itself from the live transcript. Every action is timestamped so any past graph state can be replayed or branched from. The map is the summary — there is no end-of-session AI write-up.
| Folder | Owner | Stack |
|---|---|---|
backend/ |
FastAPI + Motor + WebSocket relays | Python 3.11+ |
transcript/ |
Browser-direct mic → ElevenLabs Scribe v2 (Web Speech fallback) | TypeScript |
agents/ |
Two uagents processes (topology + enrichment) on Groq Llama 3.3 70B |
Python 3.11+ |
frontend/ |
React + Vite + reactflow + Framer Motion | TypeScript strict |
shared/ |
Frozen integration contracts (Pydantic + TS types + uAgent models) | — |
See shared/README.md for the contracts that bind these together, and
INTEGRATION.md (after Phase 2) for any contract drift that surfaced during
the build.
No paid LLM SaaS. Groq → Gemini fallback. ElevenLabs free tier → Web Speech
fallback. MongoDB Atlas M0, Cloudinary free tier, Agentverse free.
grep -r "import anthropic\|import openai" . must return nothing.
cp .env.example .env # fill in values
# Backend
(cd backend && pip install -r requirements.txt && uvicorn main:app --port 8000)
# Agents (two processes)
(cd agents && pip install -r requirements.txt && python topology_agent.py &
cd agents && python enrichment_agent.py &)
# Frontend
(cd frontend && npm install && npm run dev)