There is no paid tier, no seat count and no hosted plan. You self-host it, you own the data, and nothing is gated behind a licence key.
Everything the project does. No feature is withheld from the free version, because there is no other version.
Funds continued development, and unlocks the complete internal architecture reference that isn't published in the repository.
BrainLLM has no infrastructure of its own. The only cost is wherever you put Trilium.
Your memory should not be someone else's product.
Every note BrainLLM writes lives in a note application you control, in a format you can read without it. If this project disappeared tomorrow, your brain would still open, still be editable, and still make sense — which is the actual test of whether data is yours.
No, and there isn't a tier. The whole project is MIT-licensed on GitHub. Sponsorship buys the Blueprint — a documentation artefact — and supports the work. It doesn't unlock code.
A running TriliumNext instance, an ETAPI token, and Bun 1.0 or later. That's genuinely the hard part of setup, and it's also exactly why your data stays on your own machine.
It's a standard MCP server, so any MCP-capable client can use it. Claude is what it's exercised against daily, and the OAuth surface exists specifically so claude.ai, Claude mobile and Cowork can connect — those can't send a static bearer token.
Because you can't open one and read it. Every note here is a real note in a real editor — you can correct what the model wrote, by hand, and it will read your correction next session. That round trip is the whole point.
Only between your MCP client and your own Trilium instance. There's no BrainLLM service, no account, and no telemetry. The npm package and this site are the only things the project publishes.
Yes. MIT. Fork it, embed it, build a product on it. If it becomes load-bearing for you, sponsorship is the way to keep it maintained.
Built and maintained by one person, in production every day.