AgentDock
Download

Learn

How AgentDock works.

The short version of everything. The documentation has the long one.

Getting started

Install AgentDock, connect at least one agent you already use, and add a project folder. That’s the whole setup.

  • Download for macOS or Windows
  • Sign in to Claude Code, Codex or Antigravity with your own account
  • Choose a folder and ask for something

The workspace

A folder is a project. Every conversation in it stays with it, and every change an agent makes is there to review — or revert — beside the conversation.

  • Projects and conversations, grouped by day
  • Changes with diffs and one-click revert
  • A file explorer with previews
  • Compare agents side by side on the same question (built, not yet in the downloads)

AIgency

Describe an outcome. AIgency drafts a plan of tasks across your agents, runs independent ones side by side once you approve it, and reports on the run. Nothing is repaired without you. Part of AgentDock Pro’s developer preview.

  • A plan you edit before anything runs
  • Tasks given to Claude Code, Codex and Antigravity
  • Isolated git worktrees for parallel work
  • A report on each run; repairs only when you start them
An AIgency plan: a workflow map of five tasks, each with its agent and model.

Media

Ask @media for a launch film, a product demo, an explainer or a social cut. An agent directs the production, starting with free local tools.

  • Modes: Full Production, Ad / Launch, Social Short, Product Demo, Explainer and more
  • Local rendering with FFmpeg and native motion graphics
  • Browser services are off by default and enabled per project

Media is in AgentDock Pro’s developer preview.

A Media Production card: type, duration and format inferred from the request, with free and paid tools.

Agents and models

AgentDock runs the agent CLIs you install and sign in to yourself, and puts every model they offer in one picker, grouped by who makes it. Hand a conversation to another agent and its context goes with it.

  • Claude Code, through Anthropic’s Agent SDK
  • Codex, through OpenAI’s Codex SDK
  • Antigravity, with Google’s Gemini models
The model picker, grouped by Anthropic, OpenAI and Google.

Permissions

Every agent asks before it edits files or runs commands, and you can say no. In AgentDock Pro’s developer preview you can also choose how far each agent may go: Restricted reads and plans, Standard asks before edits and commands, Expanded edits without asking but still checks risky commands.

The Permissions menu: Restricted, Standard or Expanded, and what each allows.

Browser services

Media productions can use services like Gemini, Google Flow and Higgsfield in AgentDock’s own Chrome profile — limited to each service’s own site and off until you enable it for a project.

Part of the same developer preview as Media.

Agents that call for help

Your agents can ask for AIgency or Media themselves. The request appears in the conversation as a card with every setting visible, and nothing runs until you allow it.

In AgentDock Pro’s developer preview.

Things to try

AIgency comes with templates for common jobs. Start from one, edit the plan, and run it.

  • Build a feature
  • Fix a bug, with a tests-pass check
  • Review a repository — two independent reviews, one report
  • Plan and implement an MVP
  • Investigate a technical problem
  • Refactor code
  • Test and validate an implementation