What would you trust an AI agent to do in your notes?

I built Brain around plain Markdown files on my own server, with notes, mail and tasks in one place.

Then I connected an AI agent through MCP. Writing a note felt straightforward. Letting it send an email felt like a much bigger decision.

In Brain, you choose the permissions and can review the agent’s actions in an activity log.

Where would you draw the line? What would you let an agent do on its own, and what should always need your approval?