Platform
One workspace for the agents you already use.
AgentDock is a desktop app that runs Claude Code, Codex and Antigravity on your own computer — in projects, with approvals you control, every change reviewable, and bigger jobs handed to AIgency.
- Agents
- Claude Code, Codex, Antigravity
- Runs on
- macOS and Windows; Linux experimental
- Your data
- On your computer, no cloud sync
- Model access
- Your own CLI sign-ins
Agents and models
Three agents, each through its own engine.
AgentDock drives each agent the way its maker built it to be driven, and never sells you model access: you install and sign in to each CLI yourself. A setup assistant finds them, and can install Claude Code and Codex for you.
Claude Code
Anthropic
Runs through Anthropic's Claude Agent SDK, with live model and permission switching.
Codex
OpenAI
Runs through OpenAI's Codex SDK, with the model list read live from your own Codex install.
Antigravity
Google
Google's agent through its structured stream, with Gemini models and its own plan mode.
One model picker.
Every model your agents can launch, grouped by the company that makes it, with the agent that runs it underneath. Reasoning controls appear only for models that have more than one level. Models the catalogue marks manual-only are never picked automatically.
Projects and conversations
Work lives with the code it belongs to.
A project is a folder on your computer. Conversations, AIgency runs and Media productions are listed with it — running ones first, then by day — and titled automatically from what you asked. A built-in “No Project” space holds quick questions that don't belong anywhere.
Search across conversations, projects and runs from anywhere with ⌘K. Every conversation is stored on your computer.
Approvals and permissions
You decide what an agent may do.
When an agent wants to edit a file or run something risky, it asks in the conversation: Allow, Deny, or open the terminal. Each agent's own permission modes are available, set per conversation.
| Agent | Permission modes it offers |
|---|---|
| Claude Code | DefaultAccept editsPlanBypass permissionsDon't askAuto |
| Codex | DefaultRead-onlyWorkspace writeFull accessBypass |
| Antigravity | DefaultAccept editsPlanBypass |
Modes that skip approvals entirely are marked dangerous in the app. The next release adds Restricted, Standard and Expanded presets that map onto each agent's modes.
Around the conversation
Everything you'd open a terminal for.
Changes
Changed files and their diffs for the project, beside the conversation. Revert a file in one click.
Terminal and diagnostics
The raw CLI output for terminal-based sessions, a trace log for structured ones, or the session in your own terminal.
Continue with another agent
Hand a conversation over with a summary of what was asked, which files were touched and what's still open.
Provider usage
What each CLI reports about your plan's usage, read from the CLI itself rather than estimated.
Attachments
Attach any file. Each agent gets what it reads natively; anything else is copied into the project for it.
Saved instructions
Reusable Markdown instructions you attach to a conversation instead of pasting the same brief again.
Compare agents
Ask all three. Read the answers side by side.
Compare mode sends one question to Claude Code, Codex and Antigravity in the same project and shows the answers next to each other. Every agent runs in its read-only or planning mode, and any request to do more is declined automatically — comparing never changes your code.
Capabilities
Coming in the next releaseAgents can ask for bigger tools.
Claude Code, Codex and Antigravity can request AIgency or Media from inside a conversation — or you can, with @aigency and @media. A card shows the settings it would use; nothing runs until you allow it, once or always for that project.
Local-first
No cloud copy of your work.
Agents run as processes on your machine, against your folders. AgentDock keeps its history in a local database. The only call it makes to Sarja is to check your account and plan.
On your computer
Everything you work on.
- Projects and their folders
- Conversations and agent sessions
- Prompts and attachments
- AIgency runs and reports
- Media productions and generated files
- Each agent's own CLI sign-in
Kept in AgentDock's local database and your own folders. There is no cloud sync.
With your Sarja Account
Only what an account needs.
- Who you are (name, email)
- Your plan and entitlements
- Billing, through Stripe
- Which apps are signed in
Bring your agents. Keep your work.
AgentDock is free to download. Sign in with your Sarja Account when you want Pro — your projects and conversations stay on your computer either way.