Remote Agent Control

  • Mobile app
  • All plans
View as MarkdownFeature page

What it does

Once the phone is paired, the AgentsRoom mobile app is a full cockpit for the agents running on your computer, not a read-only viewer. From the phone you open a project, start a new agent (role and model included), read the terminal as it streams, answer a permission prompt with Allow or Deny, send a new instruction, fire a saved prompt from the Prompt Library, move a backlog ticket, duplicate or close an agent, and preview a dev server through a secure tunnel. The desktop does the work; the phone drives it. It is designed for the moments away from the desk: commute, lunch, the sofa.

Where to find it

  • Mobile app tabs: Projects (projects, agents, terminals, backlog), Activity (its Active agents segment for what happened across projects, its Usage segment for the usage gauge of the foreground agent's account, and the Companion button), Community, Settings.
  • Inside a project: the agent list with swipe actions, the agent console with its composer, its floating Agent remote and More keys grid, its header (spec strip, role pill, Agent's card), and a dock with Prompt library, Skill library and Show the conversation.
  • Pairing lives in the desktop footer (Mobile) and in the mobile Settings; see Mobile-Desktop Sync.
  • Mobile Settings has a Search all settings field and category chips (All, Desktop, Account, Notifications, Appearance, Voice, Usage, Help): typing a word (a setting name, a word of its help text) shows only the matching cards, in the phone's language.

How to use it

  1. Pair once by signing in to the same account on both, then tapping your computer in the phone's list of desktops.
  2. Open Projects, pick a project: you see the same agents as on the desktop, with their status.
  3. Tap an agent to open its console. Type in the composer to send a message; Ctrl or Alt can be armed to send a chord (Ctrl+C, Alt+B), the Agent remote sends arrows, Enter, Escape, Shift+Tab and Ctrl+C in one tap, and More keys gives the full terminal keyboard (see Terminal Keys and the Agent Console). Show the conversation switches to the chat-style transcript (Conversation View).
  4. When an agent asks for a permission, the Permission Required card appears: Allow or Deny.
  5. Start a new agent from the project screen; the desktop spawns the terminal and the session appears on both sides. Swipe an agent row to close it, or duplicate it.
  6. Live Preview: pick a running dev server (common ports are scanned, or type one), choose a subdomain and Start Preview to open it in the phone's browser through an HTTPS tunnel.

Saved prompts: open Prompt library from the dock, choose Project, Global or Team, and send a prompt to the current agent.

Usage of a given agent: the Agent's card of the console and the Usage segment of Activity show, under "By project and agent", the share of the plan window that this project and this agent consumed on the connected desktop (an estimate, described in Claude Code Token Usage).

Agents on your other computers: the Remote action of a project lists the agents, team runs and terminals of your other machines. From the phone you can attach to them and also stop an agent (Stop in the console header: "Stop this agent on <device>?"), stop a team run (the cross of its card: "The run stops on <device> and its consoles close there.") or close a terminal (the cross of its row, or Close on its screen: "Close this terminal on <device>?"). Each one asks for confirmation first. Details in Remote Fleet.

Desktop app stuck while you are away: Settings > Restart desktop (under Connection, next to Check desktop update) quits and reopens AgentsRoom on the paired computer after a confirmation. Agents and terminals are restored from the restore snapshot, the phone shows Desktop is restarting… and reconnects on its own when the desktop is back; if the desktop never answers, the banner says Desktop restart failed.

Settings

  • relayEnabled (global, Settings > Mobile): the channel the remote control runs on.
  • sendNotificationsToPhone (global, Settings > Mobile): get told when an agent finishes or needs you while you are away.

Agent tools (MCP)

None. Every action taken from the phone runs through the same desktop code as the corresponding desktop button.

Providers

All providers: Claude Code, Codex, GitHub Copilot CLI, Cursor, Antigravity and the other supported CLIs can be driven side by side from one phone session. Full-screen CLIs are mirrored as screen frames so they stay readable on the phone.

Mobile

This is the mobile app. iPhone, iPad and Android tablets (the layout scales to the larger screen); both apps are free.

Limits

  • The desktop app must be running and online. Agents keep running if the phone drops; the phone reconnects when connectivity returns.
  • Live Preview tunnels require signing in on the desktop. Free and Plus accounts get 1 tunnel start per week, Pro and Team have no quota (see Localhost Tunnel). On the phone, a Free or Plus account can also run only 1 tunnel at a time in a project: the sheet then says Tunnel limit reached and asks you to stop the running tunnel before starting a new one.
  • A project that exists only on another of your desktops is shown as Not on this desktop and cannot be started from here; switch desktop first.
  • Features that need the desktop screen (Dynamic Island, the review modal, the launchpad intent bar) have no phone equivalent.

Common questions

  • Is this SSH with a nicer UI? No. It is a native app with project views, chat-style output, typed input, tunnel controls and push notifications, over an encrypted relay, not an SSH session.
  • Does my code leave my computer? No. Agents run on your machine; the phone receives encrypted output and sends encrypted commands. Running a task elsewhere is a separate, explicit mode: Cloud Agents.
  • Can I start a brand-new agent from the phone? Yes, on any project present on the connected desktop; role and model are chosen on the phone.
  • Can I stop something that runs on another of my computers? Yes, since 1.184.0: from the Remote action of a project, stop the agent, stop the team run or close the terminal on that machine, after a confirmation. Handing off a run or answering its question is still done on the machine that runs it.
  • Does it work over 4G? Yes; only text-level data and small tunnel packets are exchanged.

Related