Agentdock

AgentDock

Stop managing AI coding agents like terminal tabs.

Mission control for parallel Claude Code and Cursor Agent sessions. Run a fleet across repos from one local-first cockpit — desktop or phone. See who is working, who needs you, review the diff, and jump in.

- Parallel Claude Code and Cursor sessions
- Isolated git worktrees
- Live working / waiting / done status
- Plans and diffs before you merge
- The same UI on your phone
- Local-first — your machine, no cloud

Install

Or clone it yourself:

./setup.sh installs Bun and tmux if they are missing, then prints a checklist. Claude or Cursor is only required when you start an agent — the UI opens without them.

Re-run the checklist anytime with agentdock doctor.

First launch asks where your repos live and which ones to add. Password, Linear, Slack, and ngrok are optional and live in Settings.

Opens http://localhost:5173 (API on port 4800).

What changes

Before

After

Three ways people use it

Run a team of tasks. Start several Claude or Cursor sessions across one repo or many. Isolated worktrees keep branches from colliding. The queue shows working / waiting / done — Claude Code hooks where they exist, terminal fallback for Cursor.

Review before you merge. Open the plan the agent wrote, read the live diff, comment, and send a follow-up without leaving the session. Restore a stopped session and the conversation comes back.

Walk away. The same UI is built for a phone. Start four jobs on a laptop, close it, answer the one that needs input from your pocket. On the LAN use https://:5173; away from home, toggle ngrok in the header.

Why this exists

I got tired of Cursor eating RAM, of not being a terminal person, and of features that span four repos. Sometimes I just need things to keep moving while walking…