Lumi is an open-source research manager where AI agents scout papers, generate hypotheses, design experiments, and write documentation — all from a pixel-art command center.
From literature surveys to paper writing, Lumi provides a complete pipeline with AI agents that do the heavy lifting.
Interactive canvas where your AI agents walk around, work at desks, and attend meetings. Watch your team in action.
Scout, Theorist, Architect, Coder, Data Smith, Documenter, and Commander — each with deep expertise powered by Claude.
Multi-agent discussions where agents debate topics in rounds. Select participants, set an agenda, watch them collaborate.
Natural language control over your entire project. Run agents, start meetings, manage entities, and get strategic advice.
7-stage research pipeline from survey through writing. Track tasks, assign agents, monitor progress at every phase.
Automatic documentation to Notion. Agent progress, meeting summaries, and project state — all synced with descriptive sub-pages.
Each agent is a specialist powered by Claude CLI. They analyze your project context and return structured insights. Create custom agents for any role you need.
Literature Research
Hypothesis Generation
Experiment Design
Implementation
Dataset Engineering
Notion Documentation
Mission Coordination
The Oracle is your AI command center. It has full control over agents, entities, meetings, and documentation. Just describe what you want — it takes action.
Lumi runs locally with SQLite. The only external requirement is the Claude CLI for AI agents.
Clone the repo and install dependencies.
git clone https://github.com/shanyang-me/lumi-research
cd lumi-research && npm install
Run Prisma migrations to create the SQLite database.
npx prisma migrate deploy
cp .env.example .env
Build and start the server. Make sure Claude CLI is installed.
npm run build
npm start -- -p 3001
Open localhost:3001, create a project, and summon your AI team.
> open http://localhost:3001