Your AI tools' shared memory, self-hosted
A Supabase-backed database and MCP gateway that lets every AI you use—Claude, ChatGPT, Cursor—draw from one persistent store of your knowledge.

What it does
Open Brain is a self-hosted “infrastructure layer for your thinking”: a PostgreSQL database with vector search, an AI gateway, and a chat capture channel, all wired together through the Model Context Protocol (MCP). You build it on Supabase, connect your AI tools, and they all read from and write to the same memory. The project includes six guided extensions—household knowledge, home maintenance, family calendar, meal planning, CRM, job hunt—that layer on top of each other and share data.
The interesting bit
The bet is on interoperability over any single app. Because everything speaks MCP, your CRM extension can reference thoughts you captured in Slack, and your meal planner can check the family calendar. The README is unusually explicit that this “isn’t a notes app”—it’s a database with an open protocol, which is either refreshingly honest or a bit of positioning theater, depending on your cynicism level.
Key highlights
- MCP-native: plugs into Claude Desktop, ChatGPT, Claude Code, Cursor, and “whatever ships next month”
- 45-minute setup guide claims no coding experience needed; there’s also an AI-assisted setup path for Cursor/Claude Code users
- Active community contributions: 20+ merged recipes for importing data from ChatGPT, Perplexity, Obsidian, X/Twitter, Instagram, Google Takeout, Gmail
- Recent PRs show real iteration: edge function cost optimization (73% invocation reduction), provenance chains, entity extraction workers, wiki synthesis pipelines
- Row-level security primitives for multi-user isolation when you get that far
Caveats
- Self-hosted means self-hosted: you’re managing Supabase, edge functions, and MCP server connections; the “no SaaS chains” pitch is also “no one else to blame when it breaks”
- The ambitious extension roadmap (household → career management) is front-loaded with beginner projects, but the advanced ones are unbuilt territory for most users
- Some recent contributions are thinly documented in the README (the “Life Engine Video” recipe trails off mid-sentence, for instance)
Verdict
Worth a weekend if you’re already juggling multiple AI tools and tired of re-explaining context to each one. Skip it if you want something that works in five minutes or don’t want to own a database.