Introduction

What is AgentDesk and why it exists.

AgentDesk is a self-hosted mission control for Claude Code agents. It provides a web-based dashboard, a proactive dispatcher, and a multi-session runtime — everything you need to orchestrate teams of AI agents that write code, manage tasks, and ship features autonomously.

Why AgentDesk?

Claude Code is powerful on its own, but managing multiple agents across multiple projects quickly becomes chaotic:

  • No visibility — agents run in terminals, outputs scroll away
  • No coordination — agents don’t know about each other’s work
  • No persistence — sessions are ephemeral, context is lost
  • No scheduling — you have to manually trigger every run

AgentDesk solves all of this with a single self-hosted platform.

Key Concepts

Projects

A project is a workspace that groups related tasks, agents, files, and schedules together. Each project has a mission statement that guides agent behavior.

Agents

Agents are Claude Code instances with persistent identities. Each agent has a name, role, custom system prompt (CLAUDE.md), and configurable LLM settings. Agents can be workers (execute tasks) or orchestrators (decompose and delegate).

Tasks

Tasks live on a Kanban board with six statuses: todo, assigned, planning, in-progress, review, and done. Tasks have priorities, subtasks, comments with @mentions, tags, and deliverable descriptions.

The Dispatcher

The proactive dispatcher is AgentDesk’s autonomous engine. It monitors the task board and automatically assigns work to idle agents, manages session lifecycles, handles errors with exponential backoff, and ensures fair distribution across agents.

Chat

Every agent has a persistent chat session. You can talk to agents in real time, see their tool calls, and collaborate on tasks — all through the web UI.

What’s Next?