agent-desk

agent
Security Audit
Warn
Health Warn
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local agent workspace with coordinators, terminals, task automation, and annotation feedback

README.md

Agent Desk
Run, coordinate, and review coding agents from one browser tab

Checks Python 3.11+ Node.js 18+ macOS or Linux License: Apache 2.0

Agent Desk is a local workspace for coding agents. Run any coding agent side by
side in real terminals, with built-in support for Claude Code, Codex, Grok,
OpenCode, and Pi, and a custom command option for any other agent CLI.
Hand work to a coordinator that splits it across agents, and review what comes
back in one inbox.

Agent Desk demo: three Claude Code and Codex sessions are open side by side. One opens a code review of its change inside its own pane. The review is opened full screen, a comment is left on a line of the diff and sent to that agent, and the pane goes back to the grid. While the agent edits the code, its pane is resized from the corner and dragged to the other side, and the review reloads with the new diff. The sessions are grouped in the sidebar. Then on a new desk a coordinator is asked to add a password rule, launches three Claude Code and Codex workers whose terminals open beside it, and reports back when they finish. Last, an automation is scheduled for weekdays

✨ What you get

  • 🖥️ Terminals for Claude Code, Codex, Grok, OpenCode, Pi, and more, up to 16 in resizable layouts
  • 🗂️ Desks that keep separate projects' agents apart, each with its own panes
  • 🧭 Coordinators that hand work to agents and track what comes back
  • 🔀 Model router support, so coordinators pick each worker's model and harness per task
  • 🗺️ Agent map showing which agent launched which
  • 📥 Inbox for questions, finished work, and scheduled runs
  • 📌 Annotations you click onto the UI and send to an agent
  • 🧩 Skills shared across every agent you use

🚀 Quick start

The quickest way in is to let your coding agent set it up. Clone the repo, open
it in Claude Code, Codex, or Grok, and say:

Set up Agent Desk with the skill in skills/agent-desk-setup.

The agent-desk-setup skill checks the
prerequisites, starts Agent Desk, and adds your Claude and Codex accounts.

To do it by hand:

git clone https://github.com/dylan-murray/agent-desk.git
cd agent-desk
python3 -m agent_desk

Then open localhost:4310. Pass --port to use another port.

Requires macOS or Linux, Python 3.11+, Node.js 18+ (22+ to run the test suite), and the agent CLIs you want to use. To install the agent-desk command instead, run pip install . from the checkout.

Using more than one Claude or Codex login? Add each under New → Terminal → Accounts.

Agent Desk only listens on 127.0.0.1. It saves its state in data/, or wherever AGENT_DESK_DATA_DIR points.

📚 Docs

Guide · Contributing · All docs

📄 License

Apache-2.0. Bundled xterm.js and highlight.js keep their own licenses in agent_desk/static/vendor/.

Reviews (0)

No results found