trusty-mpm

One binary, one daemon, many sessions

Running one coding session is easy. Running five across three repositories, remembering which worktree each belongs to and which are still waiting on you, is the part that goes wrong. trusty-mpm is the process that keeps that straight.

Everything ships as a single binary installed under two names — tm and trusty-mpm — with each surface behind a subcommand rather than a separate package to install and version-match.

• tm daemon — the background service everything else talks to.
• tm tui — a terminal dashboard across every live session.
• tm telegram and tm slack — remote control from a phone when you are not at the terminal.
• tm gui — an optional desktop shell over the same daemon.

Sessions you can walk away from

tm launch provisions a session with everything it needs already deployed — instructions, agent roster, skills — and starts it. tm ls and tm f find one again by name or by prefix; tm attach reconnects. The daemon holds the roster, so closing a terminal does not lose the session behind it.

Projects are registered rather than inferred, which is what makes the rest work: the same directory resolves to the same project every time, across sessions and across restarts.

Hooked into the session, not beside it

tm hook handles Claude Code lifecycle events — before a tool runs, after it runs, and when a session stops — and relays them to the daemon. That event feed is what makes the dashboards live rather than a periodic poll, and it is readable directly with tm events.

An MCP server exposes the same orchestration surface to a session itself, so an agent can enumerate sessions, delegate work, and read project state through tool calls rather than by shelling out.

When it goes wrong

tm doctor runs a full diagnostic of the stack. tm validate checks a workspace's deployed agents, skills, and settings against the canonical roster, and tm repair recovers from a deploy state that has drifted. tm health reports daemon reachability and a fleet summary in one line.