You kick off three agents on the Mac at home, then head to the office. Somewhere on the way one of them asks for permission to run a command, and it waits. The other two finish and wait too. Nothing is broken. Everything is just stopped until you're back at the desk that is eight kilometres away.

PonyMux 0.16 adds Remote Control for that. Your agents keep running on your computer. Open my.ponymux.com on your phone or any browser, and you see the same agents with the same status lights, whether they are Claude Code, Codex, or Grok. Open the waiting one, choose Start control, and answer it. It's free, on macOS and Windows.

How do I turn it on?

Select the cloud next to Agents in the sidebar, then Sign In…. PonyMux opens your browser, you sign in with GitHub or Google, and confirm that this computer may share. That's the whole setup. The card then shows a QR code: scan it with your phone and the web app opens with your agents already listed.

The Remote Control card in PonyMux: its state is On, with a QR code and the address my.ponymux.com, opened from the cloud next to Agents in the sidebar.

Each computer signs in once and reconnects on its own after that. Agents are reachable only while the computer is awake, so Keep Awake is worth turning on if you plan to leave.

What can I do from the browser?

Two things, kept apart on purpose.

Watching is the default. Open an agent and you see its Terminal, refreshed about once a second, with the Inspector beside it: working directory, model, pull requests, and the files it made, which you can download. Nothing you do there reaches the agent.

my.ponymux.com in a browser: the list of agents on the left, the selected agent's Terminal waiting for an answer in the middle, and the Inspector with working directory, pull requests and files on the right.

Controlling starts when you choose Start control. Then your keys go to the real Terminal: arrows, Esc, Ctrl combinations, all of it. You can also paste, drag, or pick an image or file to send, and change that one Terminal's text size. On a phone, a key bar adds esc, tab, ctrl, and the arrows that the keyboard lacks.

Opening an agent is a look. Starting control is a decision.

What if two devices want the same agent?

One device controls an agent at a time. The other one sees who has it, for example iPhone in control, and what it did last, like typed 12 chars. It can take over, and the first device goes back to viewing and is told.

On the computer, the Terminal's title bar shows who is controlling it, and a row under the toolbar has Stop. That takes it back at any time. It's the one thing that always works from where the agent actually runs.

The PonyMux window while a phone controls an agent: an iPhone icon in the title bar, a Stop button, and the Remote Control panel listing who is controlling it.

Is it safe to type into my computer from a browser?

I thought about this more than about anything else in the feature. Only your account can see and control your agents. The server relays the screen, your keystrokes, and files, and saves none of them. Before each key, PonyMux on the computer checks that the agent is still the program in front and still running, and drops the key if not, so a key doesn't land in the shell under an agent that just exited.

One gap remains. If the agent exits in the exact instant a key is written, a single Enter could reach the shell. End control when the agent is done, and use Stop if something looks off. How Remote Control stays safe has the full list, including what it doesn't have: no end-to-end encryption, and no approval cards that answer for you.

Does it work for Claude Code, Codex, and Grok together?

Yes, and that's the part I care about most. PonyMux runs the real CLI in each Terminal, so Remote Control doesn't need each agent to offer its own remote. Every agent running in PonyMux shows up in the same list, with the same status lights, whichever one it is.

The agents' own remotes work differently. Claude Code has /remote-control, which you run in the session you want to reach. Unless you turn on auto-connect, a session you didn't start it in can't be reached, and outside server mode each instance supports one remote session at a time. It needs a Pro, Max, Ultra, Team, or Enterprise plan, and API keys aren't supported (docs). Codex has its own: the ChatGPT mobile app connects to the ChatGPT desktop app on your Mac or Windows PC (docs). For Grok Build, xAI's web and mobile versions are hosted, and I haven't found an official way to reach a session on your own computer; what exists is community tools. Those are fine if you use one agent. With three, it's three setups and three apps.

The agent changes. The place you answer it doesn't.

What does it cost?

Remote Control is free. On Free, one computer shares its agents at a time. If you want more than one sharing together, Pro is coming soon, and nothing else in Remote Control is held back on Free. Plans and limits has the details.

What else is in 0.16?

The sidebar collapses to a slim icon rail, and hovering over it peeks it open. Agent Integrations moved to the bottom of the sidebar, one line per agent. You can resume an archived or trashed Terminal in one click, create a Terminal straight into any List, and wake every suspended agent with Wake All. ⌘+, ⌘−, and ⌘0 change one Terminal's text size without touching the others. The Windows app got the same Remote Control in its own 0.16.0.

For setup, start with Remote Control, and Remote Control on your phone covers the key bar and notifications. The rest is in the changelog.

Download PonyMux, free for macOS and Windows: ponymux.com/download