Your first 10 minutes with ModelBrain

If you haven't already: install ModelBrain, then connect it to Claude Desktop, Claude Code, or Cursor. Each is a couple of clicks, no config files to hand-edit.

ModelBrain starts empty — it only knows what you tell it or what you point it at. Open ModelBrain's Settings and add a folder of notes under Connectors (Markdown, TXT, PDF and DOCX are read), or from the command line:

modelbrain connector_add ~/Documents/Notes

It only reads folders you explicitly approve — see the docs for the full connector command set.

In your assistant, just talk normally:

"Remember that we moved the launch to next Tuesday because of the notarization delay."

…later, a new conversation…

"When's the launch, and why did it move?"

A working setup answers the second question correctly, in a new conversation, with no re-explaining. That's the whole product in one exchange.

The more interesting test is a question no single note answers by itself — two-hop reasoning, stitching two separate facts together. For example, if one note says who owns a project and another says that person is out this week, ask "who should I ping about the project right now?" A citation-verified answer names both notes it drew on. If it can't verify an answer from what's actually stored, it says so instead of guessing — that's deliberate, not a bug.

Ask your assistant to forget the thing you just told it, then ask about it again in a new conversation. It should come back empty. See what a forget call actually has to guarantee for what "forgotten" really means here.