Get started with antgrid

antgrid is the control plane for the coding agents you already use. Setup starts on desktop: install the app, open a project and run your agent. When you're ready, tell Handler what's left and when to ask you. Connecting your phone is optional. Free while in beta, no card required.

Before you start: bring a coding agent

antgrid drives a terminal coding agent — it doesn't ship one. Install at least one on the computer where you'll work and make sure it runs from a plain terminal:

Step 1 — Install the desktop app

Everything runs inside one desktop app — your agents' terminals, files, git and browser preview. There's no separate CLI or background service to install.

Windows

Download for Windows fetches the Microsoft Store installer. Run it, then launch antgrid from the Start menu.

macOS

Download the .dmg, open it, drag antgrid into Applications, then launch it.

Linux

Download the AppImage, make it executable, then run it:

chmod +x antgrid-linux.AppImage
./antgrid-linux.AppImage

All desktop builds are published onour GitHub releases.

Launch the app and sign in — email magic link, GitHub or Google, or Sign in with Apple on a Mac. That account is what links your devices together.

Step 2 — Open your first project

  1. In the desktop app, open the folder of a project you're working on.
  2. Open a terminal in it and start your agent exactly how you do now —claude, codex, or whatever you use. antgrid wraps the real terminal, so nothing about your workflow changes.
  3. antgrid recognises Claude Code, Codex, opencode, Cursor and other common agents for session naming and notifications; anything else still runs as a plain terminal.

Try Handler on your first task

Handler takes on the follow-ups when a task runs long and you're ready to step away. Start with a small handoff so you can see how it responds.

  1. Start a task with your agent and work through as much as you want.
  2. Tell Handler what's left, how to proceed and when to ask you. Arm it on that running session.
  3. Watch how it checks the agent's responses against your instructions and asks for evidence before moving on. Review its next instruction.

For example, after researching and planning a small feature with your agent, adapt these instructions to your project:

Implement the agreed plan and show passing test output. Then review the code, followed by a security review. Report findings with file references and fix issues within scope. Compact context between phases, preserving the plan and remaining instructions. Finally, smoke-test the feature and show the results. Ask me before starting the security review or going outside the plan. Stop with a summary of checks and unresolved issues. Don't commit or deploy.

Handler asks the agent for missing details or evidence and notifies you when it needs your input or you've told it to check with you. Work that needs your answer waits; other listed work can continue. Its judgement can be wrong, so review its decisions. Use commands your agent supports.

Handler is included free during beta. Your agent provider's usage charges still apply. You don't need a phone to try it.

Optional: connect your phone

The mobile apps are in beta — iOS via TestFlight, Android via Google Play internal testing. Emailcontact@radhaai.com for an invite.

  1. Install the antgrid app on your phone andsign in with the same account. Your computer appears automatically — there's no pairing step or QR code.
  2. On the computer, turn on the Remote switch — the Remote chip in antgrid's title bar. It's off by default on every machine and is the machine-wide gate for remote access. Free during the beta.
  3. That's it. Live terminal output, files and sessions from anywhere, end-to-end encrypted between your devices. Our servers never see your code, prompts or terminal output.

If something doesn't work