Your machines. Your agents. One control plane.

Run your coding agents on your own machines. Follow their sessions, review changes, and send instructions in one app. Start with one machine and add others when you need them.

Work with your agent, then tell Handler what's left and when to ask you.

New Session
Projects
This machine
demo-shop (sample)
Add checkout validation
Fix flaky cart test
macbook-pro
api
web
prod-box
Sign in
demo-shop /Add checkout validation
demo-shop $ claude
> The checkout endpoint accepts empty carts and
malformed emails. Add validation and a test.
Read(src/checkout.ts)
⎿ 84 lines
Update(src/checkout.ts)
⎿ +24 -3
Write(tests/checkout.test.ts)
⎿ 41 lines
Checkout now refuses an empty cart, a malformed
email and a non-positive total, each a typed
CheckoutError. Tests pass.
>
PreviewFilesGit3TerminalsHandler
Changes+71-4Commit (1)
Msrc/checkout.ts+24-3
@@ -1,10 +1,24 @@
1-import { Cart } from './cart';
1+import { Cart, cartTotal } from './cart';
22
33export type CheckoutInput = {
44 cart: Cart;
55 email: string;
6+ couponCode?: string;
67};
9+export class CheckoutError extends Error {
10+ constructor(readonly field: string, message: string) {
Works withClaude CodeCodexopencodeCursorGitHub CopilotAntigravityKiloKimiMistral Vibe

Any other agent that runs in a terminal works too.

Long-running task? Hand over the follow-ups.

Start a task with your coding agent. When you're ready to step away, tell Handler what's left, how to proceed and when to ask you. It checks the agent's responses against your instructions, asks for missing evidence and guides the remaining work.

Research and plan together. Let Handler guide implementation and reviews, and notify you when it needs your input or reaches a checkpoint you set.

Handler works inside a session you arm. Review its decisions; its judgement can be wrong.

Try one handoff

An example handoff

See which session needs you.

Every session on every machine, with its current status, in one list.

New Session
Projects
studio-workstation
relay
Needs you — which migr…
app
macbook-pro
api
web
prod-box
evals
infra
Sessions · 6 total1 needs you
studio-workstation · 2
Needs you — which migration strategy?just now
Built release bundle12m
macbook-pro · 2
Refactor auth middleware2m
Write checkout tests5m
prod-box · 2
Run the E2E suite1m
Tail deploy logs3m
Describe a task or ask a question

Several agents running. One needs a decision. Find it without opening every terminal.

Run agents where your code already is.

Your agent runs on your machine, with your existing repo, local database and development tools. Connect through antgrid without setting up a VPN or port forwarding.

Away from your desk? Check the session running on your workstation and reply from your phone.

Connect from your phone

  1. 1Install antgrid on your machine and phone
  2. 2Sign in to the same account on both
  3. 3Enable remote access on the machine

Both devices connect out to the relay. No inbound port required.

Give the session its own branch.

Start an isolated session in a Git repository. antgrid creates a separate checkout and branch for it. The session's files, diffs, terminal and preview follow that checkout, so you can review its changes before merging.

Let an agent work on a fix while you keep working in your own checkout.

Two isolated sessions, illustrated

  • auth-tokensClaude Coderunning in its own checkout
  • checkout-testsCodexready for your review

Separate working files. Not a sandbox.

One agent's finding. Another agent's next step.

Your Claude Code and Codex sessions can share findings and ask each other questions while working on the same repository.

One agent changes an API. It tells the agent writing the tests what changed, without you copying messages between terminals.

An exchange, illustrated

  • auth-tokensClaude Codelaptop

    The refresh endpoint now returns expiresAt. Update the test expectations.

  • checkout-testsCodexdesktop

    Is expiresAt a Unix timestamp or an ISO string?

  • auth-tokensClaude Codelaptop

    ISO 8601, in UTC.

Across machines, keep the sending desktop app open and remote access enabled on the receiving machine.

The relay can't read your session.

Terminal output, prompts and file contents are encrypted between your machine and your connected devices. The relay forwards them without the keys to read them.

What you seedecrypted
studio-workstation · Claude Code
02:14  Editing auth/session.ts
       + verifyPairApproval(sig, uuid)
 142 files typechecked
? approve destructive migration?
What the relay seesciphertext
k7Vd2+Rm9pQ1fX0aZ
hN8wYc4Lbe6Tj/Ou2
gP5sKi0nDqWx3Hr+M
zA9vCt7Bf1Ul6Ey4o
Js2Rn8Xa0Pd5Mk3Iq
4Wb7Lz1Tv9Co6Ge+H
k7Vd2+Rm9pQ1fX0aZ
hN8wYc4Lbe6Tj/Ou2
gP5sKi0nDqWx3Hr+M
zA9vCt7Bf1Ul6Ey4o
Js2Rn8Xa0Pd5Mk3Iq
4Wb7Lz1Tv9Co6Ge+H
How encryption works

Start with one session.

  1. Install on your machine

    Download the desktop app for Windows, macOS or Linux. Open a project on that machine.

  2. Start a session

    Run your installed coding agent in antgrid's terminal, using your existing agent account and prompts.

  3. Hand over when you're ready

    Tell Handler what's left, how to proceed and when to ask you. Arm it, then review its next decision.

For your first run, research and plan a new feature with your agent. Then ask Handler to guide it through implementation, code review, security review and smoke tests, compacting context between steps.

Try one workflow. Watch the handoff.

Free during beta, with Handler included. Use your existing repo and agent account.

No card required. Your agent provider's usage charges or limits still apply.

All builds

Windows, macOS, Linux

First time here?

Follow the setup guide to start a session.

On your phone

Access is by invite during beta. Request iOS or Android access.