Frequently asked questions

ace is an app for running coding agents on Mac, Windows and Linux, with companion apps for iPhone and iPad, and Android coming soon. It starts the agent CLIs you already have installed, such as Claude Code, Codex and Cursor, gives each task its own thread, optionally in its own git worktree, and shows which threads are working, waiting or need you.

No. ace runs each agent through its own logged-in CLI, so your existing subscriptions and logins keep working. ace never asks for, stores or proxies provider credentials.

Yes. Add several Claude, Codex, Cursor, OpenCode, Antigravity or Pi accounts, each signed in through the CLI's own login, and pick one per thread. When one hits its limit, move the thread to another.

Claude Code, Codex, Cursor, OpenCode, Antigravity and Pi are built in. Any agent that speaks the Agent Client Protocol (ACP) also works, including Gemini CLI, GitHub Copilot, goose, Amp and Qwen Code.

Yes. Agents drive ace's built-in browser, use desktop apps in the background through computer use, and run iOS Simulators and Android emulators from the devices hub. You watch each one live and can take over or stop it.

ace runs an MCP server for every agent session, so Claude Code, Codex, Cursor and the others all get the same tools: start and wait on other agents, drive the browser, apps and simulators, show pages and images in the thread, and schedule automations. Your CLI's own permission mode decides what runs.

Fleet links your machines and devices. Linked machines let agents start threads and use each other's browsers, simulators and emulators, and you can continue a thread on another machine, uncommitted changes included. Paired devices (ace for iPhone and iPad, or a browser) follow and steer agents. Machines connect directly over your local network or Tailscale, or from anywhere through ace's end-to-end encrypted relay when you sign in to an ace account.

ace runs on your own machines; your agents and code never run on our servers. Your code goes only where your agent CLI already sends it. Devices reach your machines directly over your local network or Tailscale, or through ace's relay, which is end-to-end encrypted and can't read what it forwards. You can also host the relay yourself.

No. ace works without one. An account lets every device you sign in reach every machine on it from anywhere, through ace's end-to-end encrypted relay. It holds your email, name and a short record of each machine, never your code, prompts or provider credentials.

Yes. Start each thread in its own git worktree and the agents never overwrite each other's files. You review each thread's diff, comment on lines, and commit and push from the same panel.

The desktop app runs on macOS, Windows and Linux and updates itself. ace for iPhone and iPad pairs with your machines so you can follow, approve and steer agents from anywhere; the agents themselves run on your computers. Android is coming soon.

Yes. ace is open source under the Apache-2.0 license. You pay only for the agent subscriptions you already have.

Yes. In ace, any agent can start another on any installed provider with the delegate tool. Claude Code can hand a task to Codex, Cursor, OpenCode or any ACP agent and wait for the result.

Yes. A thread isn't done until every helper below it has finished, and it shows as needing you whenever any of them waits on you.

Yes. With linked machines, delegate takes a machine and a project, and the helper runs there as an ordinary thread. If that machine is busy, it waits in the machine's queue.

A thread can run up to 64 helpers, up to 8 levels deep. You can cap how many run at once in Settings; the rest queue.

No. Each helper runs through its provider's own logged-in CLI, on the subscription you already have.

Yes. Agents act through accessibility in the background, so they don't move your pointer or type into the app you are using. Each session shows in ace with Take over and Stop.

Yes. Choose the apps agents can't use, and ace refuses them whatever an agent asks. Computer use stays off until you turn it on.

Every agent ace runs: Claude Code, Codex, Cursor, OpenCode, Antigravity, Pi and any ACP agent. They all get the same MCP tools.

Yes. Simulators, emulators and browsers on a linked machine are available to agents on the others, as far as that machine's access setting allows, and you approve each device the same way as locally.

Yes. Linked machines let agents start threads, delegate to other agents and open browsers on each other, as far as each machine's access setting allows.

Yes. Continue on another machine takes the project and branch, uncommitted changes included, and starts a thread there from this one's history.

Directly over your local network or Tailscale when they can. From anywhere else, through ace's end-to-end encrypted relay once they're signed in to your ace account, or through a relay you host. Devices join only when you pair them or sign them in.

No. Agents run on your computers. The iPhone and iPad apps follow and steer them: approve, answer, review, browse and drive devices, across every paired machine at once.

Android is coming soon. Until then, any browser can pair with your machines through the web app.

Sign in to your ace account and it reaches every machine on it, from anywhere, through ace's end-to-end encrypted relay. Or pair it with a link or QR code over your local network or Tailscale, choosing view only or view and act, and for how long.

No. An account is for reaching your machines through ace's relay from anywhere. Everything else works without one, including pairing over your local network or Tailscale and running your own relay.

No. Traffic between your devices and your machines is end-to-end encrypted. The relay sees addresses, timing and the size of encrypted messages, and keeps none of it after the connection ends.

Your account id, email and name, and for each machine you sign in, its name, platform, key fingerprint and when it was last seen. Never your code, prompts or provider credentials.

They could open machines that have no machine code. Set a code on the machines that matter, use multi-factor sign-in, and sign out devices you no longer use.

Yes. Add each account, sign in through Claude Code's own login, and pick the account per thread or set a default. The same works for Codex, Cursor, OpenCode, Antigravity and Pi.

Your messages are held, not lost. Resume at the reset, snooze the thread until then, or move it to another account of the same provider.

ace's own browser, a real Chromium you can watch and take over, from your desktop or your phone. With Use my Chrome, agents can also work in tabs you share from your own Chrome.

Yes. The browser tools return the page's console and network log, so an agent can find and fix the error it caused.

Yes. On a Mac with Xcode, an agent can start an iOS Simulator, read its UI tree, tap and type, and read the logs. You approve the device first.

Yes. Android emulators run in the same hub with the same tools, so any agent can test an Android build.

Yes. Each thread can run in its own git worktree, so agents never overwrite each other's files, and you review each thread's diff separately.

Its turn ended, but a subagent or a background command it started is still running. The thread says Done only when all of them finish.