// the find
NateBJones-Projects/OB1
Open Brain — The infrastructure layer for your thinking. One database, one AI gateway, one chat channel — any AI plugs in. No middleware, no SaaS.
Open Brain is a self-hosted personal knowledge layer built on Supabase (Postgres + pgvector) that gives AI tools — Claude, ChatGPT, Cursor, whatever — a shared persistent memory via MCP. The core idea is one database that all your AI clients read from and write to, rather than each tool having its own isolated context. Aimed at developers and power users willing to set up and maintain their own Supabase project.
The MCP integration is the right architectural bet — rather than building yet another chat interface, it makes existing AI tools smarter by giving them a shared retrieval layer. The community contribution model is working: 20+ merged PRs with recipes covering Obsidian, Gmail, Twitter, ChatGPT exports, and more, which means the data ingestion problem is mostly solved. The decision to use pgvector directly in Supabase avoids the operational complexity of a separate vector database, and the Edge Function deployment model keeps it serverless-friendly. The automated PR review agent (checking SQL safety, secrets, structure) before human review is a smart quality gate for a community-driven repo.
The setup is a 45-minute manual process involving Supabase config, edge function deployments, and MCP server wiring — that's a significant barrier and the video walkthrough suggests they know prose docs alone aren't enough. The FSL-1.1-MIT license is not open source: it converts to MIT after two years, but until then you cannot use this commercially or compete with the maintainer, which limits adoption in team or SaaS contexts. The 'extensions' learning path mixes personal-knowledge tooling with household management and meal planning, which dilutes the core value proposition and makes it harder to explain what this is for. There's no clear sync or conflict resolution story for when multiple AI clients write to the same thoughts table simultaneously — the schema-aware routing recipe patches around this but it's not a first-class feature.