Mission control for your AI dev team.
Commander is a desktop operator workspace: plan the work, dispatch it to a fleet of coding agents, watch every terminal live — and let nothing merge without your sign-off.
The problem
Once you run more than one capable coding agent, the bottleneck moves: it is no longer writing code, it is supervising the agents that write it. Work happens in parallel, terminals scroll past, and the human loses the thread — exactly when their judgment matters most.
What Commander does
Commander puts a single operator surface over the whole fleet. You decompose a feature into tasks on a board, dispatch each task into a live embedded terminal, and follow every agent's progress in real time — Kanban, table, or dependency graph.
External agents are first-class citizens: Claude Code, Codex, Cursor, Pi, and Gemini connect through Commander's MCP server and the cmd CLI, reading and writing the same task store you see on screen.
What makes it different
- The completion gate. Factory sessions merge agent work into an integration branch — never into main. Every session ends in a human review, by design.
- A CLI as capable as the GUI. Every operator action exists as a
cmdcommand, because agents are users too. - Live, visible dispatch. Tasks run in real embedded terminals with reservation tokens and recovery — no invisible background magic.
- Local voice, local specs. Task authoring with on-device whisper.cpp dictation and built-in spec wizards across four frameworks.
- An agent-navigable codebase. 101 architecture decision records, a domain glossary, and a strict module contract — written to be edited by the agents it commands.
Install
Download the notarized DMG and drag Commander to Applications. The app keeps itself current with a built-in auto-updater.
# or grab it from the releases feed
open https://github.com/ruizrica/commander-v2/releases/latest