A markdown knowledge brain for AI agents and humans. Plain files on disk, no database, no lock-in. Works with Hermes Agent, OpenClaw, and Claude Code today. Now in alpha.
BizerBrain is a knowledge base your AI agent does most of the work on. You talk to your agent the way you normally would. The agent captures meetings, conversations, decisions, and ideas as clean structured notes. It links recurring people, projects, and topics so your knowledge base becomes a graph instead of a pile. It pulls back what you need when you ask. It keeps the structure organized over time.
Everything the agent captures is stored as plain markdown files in a folder. You can open them in any text editor. You can grep them. You can back them up by copying the folder. There is no proprietary format and no database. If BizerBrain disappears tomorrow, your knowledge is still there, intact and readable.
A web UI ships in the same container so you can drop in any time to read what the agent captured, fix something it got wrong, write something directly, or browse how your notes connect. The agent and the UI read and write the same files — there’s no syncing because there’s nothing to sync.
The agent integration works across multiple agent frameworks today. The same brain serves a Hermes Agent setup, an OpenClaw setup, or a Claude Code setup with the same skill. Your accumulated knowledge moves with you across agent and model changes, because the brain is independent of either.
Alpha is open
BizerBrain is a public alpha. The repository is at github.com/kelsi-bizer/bizerbrain and runs as a single Docker container on any VM.
If you run it and find something broken, want a feature, or have feedback on how it fits how you work, send it to feedback@bizer.ai. Alpha-stage software has rough edges and the only way they get sanded down is through people actually using it and reporting what they hit.
A few things to know going in:
- The brain works best at small to medium scale right now — hundreds to a few thousand notes. That’s plenty for personal use during alpha.
- The agent will sometimes get things wrong. That’s why the web UI exists. Fixing things by hand is part of using it.
- The underlying format — plain markdown in a folder — is stable. The app around it may change as the alpha progresses.
Why it’s shaped this way
BizerBrain is the first reference implementation of a longer architectural argument laid out in A Cognitive Architecture for AI Agent Memory — a whitepaper on what AI agent memory should actually be. The short version: memory shouldn’t die when the agent changes, when the model changes, or when the framework changes. The knowledge is the durable thing. Everything else is interchangeable.
If you want the full reasoning before you run anything, the whitepaper is the place to start. If you’d rather just try it, the repo is ready.
Bug reports, feedback, and questions: feedback@bizer.ai