A self-hosted experiment registry and persistent memory system I built for autonomous research agents.
Technologies
- Python:
- Next.js
- React
- TypeScript
- PostgreSQL
- pgvector
- Docker
Why build an experiment registry?
Research agents can repeat failed ideas when results live only in logs or chat history. Mono keeps hypotheses, code metadata, live metrics, artifacts, outcomes, and conclusions in one project so later runs can search and reuse that evidence.
How do agents use it?
Agents can work through the HTTP API, Python SDK, command-line interface, or MCP server. Mono also coordinates experiment proposals and worker claims, streams live metrics, and records the Git state attached to each run. The web app gives people the same records through a dashboard.