Documentation / Start with a task
What do you want to do?
Get one agent connected first. Then add continuity, shared work and human decisions as you need them.
Get to your first useful session
Brains is a local coordination service, not another coding agent. Your harness still writes code and talks to its model provider. Brains stores the work those sessions share in SQLite and exposes it through the CLI, MCP and a browser console.
- Install and start
Set up an isolated Python installation, register your project and check the foreground stack before opting into a background service.
- Connect one agent
Wire a supported harness, inspect its MCP configuration and understand which credentials and mailbox proofs stay private.
- Resume real context
Use a Session, search recorded knowledge and leave a checkpoint and handoff so the next agent does not start cold.
Add coordination when work is shared
Start with a clear scope, a named owner and an observable result. A claim records ownership; it is not an operating-system lock. Agents must honor it and release it when work stops.
- Share work safely
Give each agent its own Session. Claim tasks before editing, ask a peer for help and accept the exact request you reviewed.
- Ask a human
File decisions separately from peer questions. Keep working on unrelated tasks while a human reviews the ask.
- Maintain your install
Check readiness, update the package, inspect backups and review the limits of local governance before relying on it.
Guides and exact reference
These pages explain common tasks without duplicating the entire source documentation. Use the reference hub for command families and tool names, or read the release-tagged full guide and architecture for deeper context.
The supported product is alpha software for one local operator. Model routing, semantic indexing, chat bridges and runtime execution are outside this site's product contract.