OB1, also called Open Brain, is a personal knowledge database that several AI tools share as one memory. It combines a Postgres database with vector search, an AI gateway, and an open protocol so that Claude, ChatGPT, Cursor, a coding agent, or any future client reads and writes the same store instead of each keeping separate notes. The base unit is a thoughts table with embeddings, not a notes application.
The starting system is built from a Supabase project, an AI gateway, a Slack capture path, and an MCP server, deployed as edge functions and reached from any MCP client. A guided setup walks through the build, and an AI-assisted path lets a coding agent do it by reading the repository.
The repository is organised as a learning path. Six extensions build in order — a household knowledge base, a home maintenance tracker, a family calendar, meal planning, a professional CRM, and a job-hunt pipeline — and each one adds a schema and an edge function on top of the shared thoughts store, so later extensions can query earlier ones. Five primitives cover the concepts that repeat: deploying an edge function, connecting a remote MCP client, troubleshooting, row-level security for multi-user isolation, and sharing a scoped MCP server with other people.
Alongside the curated path, the repository carries reviewed community contributions in five folders: recipes that import a data source or add a workflow, agent skill packs, dashboard front ends, integrations, and database schema extensions. Import recipes exist for ChatGPT, Perplexity, Obsidian vaults, X, Instagram, Google Takeout, Grok, Blogger archives, Readwise, and Gmail. Every contribution carries its own instructions and a metadata.json file, and pull requests pass an automated review agent that checks structure, secrets, SQL safety, dependencies, and documentation before a maintainer reviews them.
Features
- Shared memory across clients: one database and MCP endpoint serve Claude Desktop, ChatGPT, Cursor, Claude Code, and other MCP clients
- Vector search: thoughts are embedded and classified on capture, so retrieval is semantic rather than keyword-only
- Edge-function deployment: every extension deploys as a Supabase Edge Function, or self-hosted on Kubernetes with PostgreSQL and pgvector
- Capture bots: Slack and Discord integrations capture messages into the store with automatic embedding and classification
- Extension path: six graded extensions that interconnect — CRM contacts, calendar, meals, home maintenance, and job applications share one brain
- Row-level security: PostgreSQL policies isolate data per user, and a shared MCP server grants others scoped access
- Data import recipes: parsing, deduplication, embedding, and ingestion for ChatGPT, Perplexity, Obsidian, X, Instagram, Google Takeout, Grok, blogs, and email archives
- Agent skill packs: plain-text skills for auto-capture, research and meeting synthesis, competitive analysis, deal memos, and financial model review
- Agent memory schema: provenance, review, use-policy, source-reference, relation, recall-trace, and audit sidecars for agent workflow memory
- Dashboards: SvelteKit and Next.js front ends that point at your own backend and can be hosted on Vercel or Netlify
- Content deduplication: write-time content fingerprints and backfill tooling remove duplicate thoughts
- Contribution review: an automated agent checks every pull request before a human maintainer merges it
