Octomux

octomux
Coding got faster. Managing agents didn't.

octomux is a local dashboard for running many Claude Code and Cursor agents in parallel. Each agent works in its own git worktree; you get one inbox for every "allow this tool?" prompt, a live grid of the whole fleet, and in-app diff review. Runs on your machine — no cloud, no telemetry, MIT.

Open localhost:7777, describe a task, pick Claude Code or Cursor, and watch it work.

What you get

Four acts, one window — from prompt to merged PR:

- ① Dispatch — Type a task (or paste a Jira/Linear/GitHub link, or a whole list). Each one gets its own worktree, branch, and agent. Pick the model and the harness per task.
- ② Run unattended — Attach a verify command and the agent loops in fresh context until that command exits 0. Put it on a cron and it does that at 3am. Hand a goal to the orchestrator and it plans, splits, and dispatches child tasks for you to approve.
- ③ Watch, or don't — Every agent's live terminal on one Monitor grid, the diff as it grows, and every "allow this tool?" prompt collected in a single inbox instead of scattered across panes.
- ④ Review & ship — An agent drafts a walkthrough and inline comments grounded on the real diff. You read it in-app, mark files reviewed, send comments back for a fix, and publish as one batched GitHub review when _you_ accept it. octomux spots the PR by branch and closes the task when it merges.

Crash, reboot, close the lid — octomux start restores every task, branch, and session.

The 60-second version

Screenshots

Every screenshot is full width — a dashboard shrunk into a table cell is texture, not information.

Review workstation — the agent opens with a verdict, a risk read, and a ranked list of things worth looking at, each linked to a line. Start there, then drop…