Run a team of agents

Unibrain is built for several agents writing to one vault, safely. Here's a setup that works well.

1. Give each agent a job and a name

Agent	Where it runs	Job
Igor	a Claude Project	research librarian: files findings carefully, with sources
Brainstorm Partner	a Claude Project	turns conversations into plans and decisions
Coding agent	Claude Code	keeps project specs, decisions and diagrams current
Hermes	your server, Telegram	quick captures, briefings, starting other jobs
Gardener	Claude Code on a timer	weekly tidy-up proposals

Each one signs its writes, so you always know who did what.

2. Agree the house rules in AGENTS.md

Put the rules all agents share in AGENTS.md. Two that work especially well for teams:

• Propose, don't act on cleanup: moving, archiving and merging happen only when you approve.
• Log big changes: keep an Agent Change Log note where each agent adds a dated line saying what it changed. Some people also keep an Agent Lounge note where agents leave messages for each other.

3. A weekly tidy-up agent with checkbox approvals

A scheduled agent keeps the vault clean without ever surprising you:

1. Sunday morning, it runs (Claude Code on a timer, for example), checks vault health, tags and folders, and writes a report note of numbered proposals, each a checkbox: - [ ] **G3** Move "Rome restaurants" from inbox/ to travel/italy/
2. You tick the proposals you like, on your phone.
3. Its next run applies only the ticked ones, signs every change, and carries the rest forward.

Tag its reports (for example #gardener) and hide that tag from Tasks in Vault settings, so its proposals don't clutter your todo list.

4. Review on your phone

• Recently changed on the home screen shows each change and who made it.
• Usage stats shows commits per agent over time.
• Ask any agent: "What did each agent change this week? Anything that needs my attention?"
• The safety net flags rollbacks and deletions, whoever caused them.