Desk

Desk is a local operator workspace for running many coding agents at once. It keeps each agent alive in a Moor session, renders the fleet in a browser multiplexer, and adds the collaboration and development tools you need around those agents: channels, an IDE, LSP-backed code intelligence, Git, GitHub, project boards, notes, and operational telemetry.

​

What Desk is for

Use Desk when you want one local control room for a group of agents working across projects, branches, terminals, files, and team-like conversations. Desk is built for:

• Agent operators who need to keep Codex, Claude, OpenCode, Qwen, Kimi, Grok, shell, and custom-command sessions alive while switching between groups quickly.
• Repository maintainers who want terminals, editor tabs, Git operations, GitHub context, project boards, and notes in one browser workspace.
• Multi-agent workflows where agents need shared channels, mentions, threads, delivery diagnostics, and explicit operator intervention points.
• Remote development boxes where code, credentials, Moor, Git, agent CLIs, and language servers already live on the host.

​

How the system is organized

Desk separates process ownership from view ownership:

• Moor owns process lifetime. Agent sessions keep running when the browser closes, the network drops, or the Desk server restarts.
• The manifest owns intent. ~/.config/desk/desk.yml defines projects, groups, sessions, layouts, startup commands, permissions, and UI settings.
• The browser owns the view. The UI subscribes to terminals, channels, files, diffs, project boards, notes, and telemetry without becoming the owner of the agent process.
• The terminal daemon owns terminal transport. One binary browser WebSocket multiplexes visible surfaces while the supervised daemon manages Moor holders, generation fencing, snapshots, input, and resize.
• Hooks own agent events. Codex, Claude, OpenCode, Qwen, Kimi, and Grok report lifecycle, attention, prompt, stop, and permission events back to Desk.

Read Architecture and Workspace model before designing a larger fleet.

​

Documentation map

​

Start here

• Getting started walks through installation, first run, agent authentication, session creation, and expected output.
• Configuration documents the manifest and settings model.
• Distribution and deployment explains source and standalone runtimes.

​

Concepts

• Architecture explains the runtime components.
• Workspace model explains projects, groups, sessions, layouts, durable session ids, and resume ids.
• Agents and terminals covers the native chat surface, terminal behavior, attention, rendering, and session controls.
• Channels protocol documents message storage, delivery, mentions, and diagnostics.

​

Guides

• Create an agent fleet builds a practical multi-agent manifest.
• Collaborate through channels shows the operator and agent messaging loop.
• Run Desk securely covers localhost defaults, SSH tunnels to your own development box, local safety, and standalone deployment.

​

Operations and reference

• Operations covers lifecycle, telemetry, attention, terminal health, session controls, and the emergency kill switch.
• Troubleshooting and FAQ maps symptoms to checks and fixes.
• API and runtime reference, Security and plugin model, Keyboard shortcuts, and Release notes provide operator reference material.

​

First decision

Install the complete CLI, then choose a server mode: