Every AI tool you use starts every conversation as a stranger. You've explained your job, your projects, your people a hundred times — and next month, when you switch tools, you'll explain it all again. Open Brain ends that: one database you own, one open protocol, and every AI you use reading and writing the same memory.
This page is the short version — what Open Brain is, how it works, and what it takes to build one (about 30 minutes, zero code). The full step-by-step guide, the extensions, and the community live in the GitHub repo, and everything here points you there.
Think about the first five minutes of every AI conversation: who you are, what the project is, what you decided last time. That's context you already gave the same tool yesterday, rebuilt by hand, again. Multiply it across every chat you open this week and the smartest tool in your life is charging you a daily tax for its amnesia.
ChatGPT's memory lives in ChatGPT. Claude's projects live in Claude. Neither can read the other, and the tool you adopt next month starts from zero. That's not an accident — a vendor that remembers you is a vendor that's hard to leave. Useful for them. Useless for you the moment you use more than one AI, which is everyone.
Strip the mystique and persistent memory is three ordinary things: a table that stores what you said, a search index that finds it by meaning, and an API any client can call. That's not a feature you wait for a vendor to ship. That's infrastructure — and infrastructure is something you can own.
Everything you capture lands in a single thoughts table: the raw text, a vector embedding (your meaning encoded as numbers, so search works by concept instead of keyword), and structured metadata. It runs on Supabase's free tier with the pgvector extension — a plain Postgres database, which means your memory is queryable, exportable, and yours in the most literal sense.
In front of the database sits an MCP server — Model Context Protocol, the open standard AI clients use to reach outside tools — deployed as a Supabase Edge Function. Claude, ChatGPT, Cursor, Claude Code, and whatever ships next month all speak MCP, so connecting a new client is pasting a URL into its connectors screen. No client is special. They're all just readers and writers of the same brain.
Because the memory lives in your database instead of a vendor's, it outlives every subscription decision you'll ever make. Cancel a tool, adopt a tool, switch models mid-year — the brain stays put and the clients rotate around it. Your context stops being a switching cost and starts being an asset that compounds.
Tell any connected AI to remember something and it writes a thought on the spot. Wire up the Slack or Discord capture and a quick message becomes a memory with no app-switching. Or go wholesale: community recipes bulk-import your Gmail archive, your full ChatGPT history, your Obsidian vault, even your X and Instagram exports. Every entry gets embedded automatically through OpenRouter on the way in.
Ask any connected client what you know and the MCP server runs a vector search over your thoughts. Ask about the kitchen renovation decision and it finds the right memory even if you captured it as a rant about contractors. The relevant context lands in the conversation you're already having — which is the whole point: the first five minutes of re-explaining just disappear.
Because every extension reads the same table, capabilities stack instead of siloing. The community's learning path takes you from a household knowledge base through a family calendar and meal planner to a professional CRM and a job-hunt pipeline — and they see each other. Your job-hunt contacts become network contacts. Your meal planner checks who's home this week. That cross-visibility is what a memory layer buys that six separate apps never will.
You need two accounts: Supabase (your database, free tier) and OpenRouter (your AI gateway — about $5 in credits, which lasts months). The guide's credential tracker keeps every key and URL straight while you build.
Eight steps, all paste-and-click: create a Supabase project, run four SQL commands (your table, your search function, your security policy, your permissions), grab an OpenRouter key, deploy the MCP server as an Edge Function, and connect your AI clients through their connector settings. If you'd rather delegate, the AI-assisted setup path points Claude Code or Cursor at the repo and lets it drive.
The moment your first client is connected, run the two-prompt test: capture in one tool, recall in another. When the second tool answers with the thing the first tool saved, you'll feel the click — that's one memory behind every AI you use.
<prompt>
<task>
In any connected client:
Remember this: I finished setting up my Open Brain today. My first goal with it is [YOUR GOAL].
Then, in a different connected client:
Search my Open Brain: what did I say my first goal was?
</task>
</prompt>The repo ships a FAQ covering the common gotchas, a Discord for real-time help, and dedicated AI assistants that know the system inside and out — a Claude Skill, a ChatGPT custom GPT, and a Gemini Gem — so the debugging help matches the tool you're already in.
Six builds in order, each teaching a concept through something you'll actually use: household knowledge, home maintenance, family calendar, meal planning, professional CRM, job-hunt pipeline. By the end your agent sees across your whole life — and you understand every table it touches.
Recipes import your existing life — Gmail, ChatGPT, Perplexity, Obsidian, X, Instagram, Google Takeout — and add workflows like daily digests and the Life Engine personal assistant. Dashboards give your brain a front end you can host on Vercel or Netlify. Skills are drop-in prompt packs for Claude Code, Codex, and friends. All community-built, all reviewed, all documented.
The Discord is where the real-time help and show-and-tell happen; the Substack carries the story and the deep dives. And when you build something worth sharing, CONTRIBUTING.md walks you through the PR — an automated review agent checks structure, secrets, and SQL safety, then a human maintainer reviews for quality. Your contribution ends up in front of the same builders who'll follow this guide tomorrow.