---
title: "Desk"
sidebarTitle: "Overview"
description: "Local mission control for coding-agent fleets."
---

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.

<Frame caption="The agent multiplexer: four durable Moor sessions in a 2x2 group">
  <img src="/images/agents-multiplexer.png" alt="The agent multiplexer: four durable Moor sessions in a 2x2 group" />
</Frame>

<Columns cols={2}>
  <Card title="Start in five minutes" icon="rocket" href="/getting-started">
    Install Desk, start the server, authenticate agent tools, and launch your
    first durable session.
  </Card>

  <Card title="Understand the model" icon="network" href="/concepts-architecture">
    Learn how Moor, the manifest, the browser, the terminal daemon, and agent
    hooks fit together.
  </Card>

  <Card title="Create an agent fleet" icon="bot" href="/guide-create-agent-fleet">
    Configure projects, groups, sessions, layouts, and agent permissions for a
    multi-agent workspace.
  </Card>

  <Card title="Operate and troubleshoot" icon="activity" href="/troubleshooting">
    Diagnose terminal, channel, GitHub, LSP, permission, and deployment issues.
  </Card>
</Columns>

## 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](/concepts-architecture) and [Workspace model](/concepts-workspace-model)
before designing a larger fleet.

## Documentation map

### Start here

- [Getting started](/getting-started) walks through installation, first run,
  agent authentication, session creation, and expected output.
- [Configuration](/configuration) documents the manifest and settings model.
- [Distribution and deployment](/distribution-deployment) explains source and
  standalone runtimes.

### Concepts

- [Architecture](/concepts-architecture) explains the runtime components.
- [Workspace model](/concepts-workspace-model) explains projects, groups,
  sessions, layouts, durable session ids, and resume ids.
- [Agents and terminals](/agents-and-terminals) covers the native chat
  surface, terminal behavior, attention, rendering, and session controls.
- [Channels protocol](/channels-protocol) documents message storage, delivery,
  mentions, and diagnostics.

### Guides

- [Create an agent fleet](/guide-create-agent-fleet) builds a practical
  multi-agent manifest.
- [Collaborate through channels](/guide-channels-collaboration) shows the
  operator and agent messaging loop.
- [Run Desk securely](/guide-deploy-securely) covers localhost defaults, SSH
  tunnels to your own development box, local safety, and standalone deployment.

### Operations and reference

- [Operations](/operations) covers lifecycle, telemetry, attention, terminal
  health, session controls, and the emergency kill switch.
- [Troubleshooting and FAQ](/troubleshooting) maps symptoms to checks and
  fixes.
- [API and runtime reference](/api-runtime-reference), [Security and plugin
  model](/security-plugin-model), [Keyboard shortcuts](/keyboard-shortcuts),
  and [Release notes](/release-notes) provide operator reference material.

## First decision

Install the complete CLI, then choose a server mode:

<Tabs>
  <Tab title="Default Bun mode">
    `desk serve` launches the release-private compiled runtime and embedded UI.
    Start with [Getting started](/getting-started), then read
    [Distribution and deployment](/distribution-deployment) and
    [Deploy and secure Desk](/guide-deploy-securely).
  </Tab>

  <Tab title="Vite development mode">
    `desk serve --dev` explicitly starts Vite from the installed source-backed
    release or a checkout. It never falls back to Bun.
  </Tab>
</Tabs>

<Note>
Desk is local-first by default. Keep it bound to `127.0.0.1`; if you work on a
remote development box, reach it through SSH port forwarding. Read [Security and
plugin model](/security-plugin-model) before adding local runtime extensions.
</Note>