None Assembly Required: Hosted LLM Wiki FTW
Nate Jones built a self-hosted AI knowledge base. It works great. It's also 45 minutes of setup you now have to maintain. Here's the hosted version.
By Ian
If you watched Nate Jones' video on Andrej Karpathy's LLM Wiki concept versus OpenBrain, you already understand the problem better than most people who've been thinking about AI for years.
The problem: every new conversation starts from zero. Your AI is brilliant and amnesiac. You re-explain your project every session. The context you've built up over months lives inside one model, on one platform, and goes nowhere.
Nate's solution is a self-hosted knowledge base. Hosted database, an MCP server, some configuration. Works great. Costs almost nothing to run.
He also filmed the setup. It's 45 minutes long.
I'm not criticizing the approach. I built the same thing — it's the basis of what brains is under the hood.
But here's what Nate says at the end of his video, after walking you through the whole setup: now you have to maintain it.
You have to keep the Supabase instance running. Manage the auth. Handle the MCP connection when it breaks. Figure out why your agent stopped writing back. Update the schema when your workflow changes.
That's the part of the video that doesn't fit in the thumbnail.
The idea Nate and Andrej are both describing is correct. An AI that writes back to a structured knowledge base, that remembers decisions, that a different AI can pick up and continue from is the right architecture. It's not a hack. It's the design.
brains is a hosted knowledge base that connects to any MCP-compatible AI. Claude, ChatGPT, Cursor, whatever drops next week. Your context lives in plain markdown files, readable by humans and machines, owned by you.
You don't configure a database. You don't write MCP server instructions. You don't maintain anything.
You sign up, connect your AI, and start. Your agent writes back automatically. Every session picks up where the last one left off.
The 45-minute setup video becomes a 5-minute onboarding.
If you got halfway through Nate's tutorial and stopped, this is for you.
Get started in 5 mins right now.