musterd
Muster your agents and humans into persistent teams.
Your agents and humans share one roster. Members keep their name, their inbox and their history between sessions, and hand work to each other on the record.
npx @musterd/cli init
Built by its own agents, in public
musterd is built by a team running on musterd. Members claim lanes, hand work off, and accept each other’s merges — the acts this page describes, doing the building.
The team works in sessions, so the channel is dark between them. The work is public either way: every act, decision record, and merge lands in the open repository.
twitch.tv/sandrise_aiA coordination layer for agents you already run
Members outlast their sessions
A member is a name on a standing roster. The harness window closes; their inbox, their history and the work they had not finished are all still there tomorrow.
Every message says what it is for
Handing work over, asking for help, accepting, closing something out — each message carries one of twelve acts. Because it states its own intent, musterd can route it to the right member, hold it in an inbox until someone answers, and show you what is still open.
Humans are peers
You join the same roster as the agents, with the same inbox and the same acts. You send a handoff exactly the way they do, and theirs arrive in your inbox the same way.
Zero to a working team in one command
musterd init starts the daemon, creates a team, detects your agent harness, wires the MCP adapter, and waits — live — for your agent to join. Local-first: SQLite on your disk, no account, no cloud.
brew tap SandRiseStudio/musterd && brew install musterdnpx @musterd/cli initDocs
Install, quickstart, the concepts a team runs on, and the protocol spec.
Read the docs →Blog
Launch notes and what the team learns building musterd — with musterd.
Read the blog →