Remote Fleet

  • Desktop, mobile and web
  • All plans

What it does

Links every computer where you are signed in to AgentsRoom into one view. The sidebar shows your local projects and, below them, the projects that live on your other machines, with live presence (online, offline, how many agents are working there). A project that exists only on another machine opens as a remote workspace; an agent of a project you have on both machines can be launched on either. The remote machine spawns the agent locally and streams its terminal back to you over the end-to-end encrypted relay; your code never moves. The same mechanism lets you invite teammates to attach to agents running on your machine. A team run started on another machine (by hand or by a trigger there) shows up as a team box too, with its steps, its counter and the machine it runs on, not as a handful of unrelated agents. Since 1.182.0 the terminals of the bottom bar (dev commands, blank shells, SSH tabs) open on your other machines appear in the project's terminal bar too, and you can follow them from here. Since 1.184.0 you can also close one of those terminals, stop a team run or stop an agent on the other machine without walking to it, from the desktop and from the phone, and a remote session looks like a local one: the bottom bar stays, the tab is colored, and a small machine badge sits at the top right of the pane. Since 2026-10-13 an agent can also MOVE to another of your computers with its conversation (Continue the conversation on): it picks up the same conversation over there, and its console here closes.

Where to find it

  • Sidebar, section On my other machines: each project says "On <device>" or "<device> is offline", with "N agents running on <device>". Since 2026-10-18 the project dot follows the agents working over there, in the same colours as a local project: red when one waits for you, orange while one works, green when the machine is online and idle, grey when it is offline. The same goes for a local project whose agents run on another of your machines, for the Shared with me rows, and for the project list of the phone.
  • Inside a project: the agent list has a Remote category ("Runs on <device>") under the cloud agents, and Add an agent on another machine.
  • Agent context menu: Run on > This computer or another online machine; Stop on <device>.
  • Agent context menu: Continue the conversation on > one of your other online computers, or This computer ("Brings its conversation here from <device>") when the agent runs elsewhere.
  • Inside a project, a team run hosted elsewhere gets its own box after the local team boxes, in both sort modes: the run title, "current / total" steps, a chevron, the machine name on the right, then a cross ("Stop this run on <device>", or "Dismiss this run on <device>" once it is finished), then the Relay or Team banner and one row per step. Click the band to fold the box down to the band, the current step and its state (the phone keeps a question or the cycles band visible too). A run its machine stopped on its max cycles keeps its box (since 2026-10-16): the band "This team used all its cycles" shows the last handoff and its report, and Add one cycle / Add two cycles resume it over there, exactly like the buttons of its own run box; the cross dismisses it (its consoles close there). Its consoles hang under the step they play (Team) or in a "Playing <step>" card (Relay).
  • Under the name of the open project (sidebar header and title bar), when the same repository also lives on another of your computers: "On <device>", always naming THIS computer, the copy you are driving. A long name is cut; its tooltip gives the full name, the local path and how many other machines hold the project.
  • Project context menu: Open on <device>, and Shared with me for projects other accounts opened to you.
  • Terminal bar at the bottom of the project: the tabs of your other online machines come after the local tabs, with the machine's name in small after the tab name. Hovering a remote tab shows a cross ("Close this terminal on <device>"). Clicking the tab opens it in the center pane, laid out exactly like a local terminal: the bare grid, the bottom bar still there, and a small floating badge at the top right reading "<tab name> · <device>" with a detach icon and a cross. When the bar holds both local and remote tabs, three icons at its far left filter it: All, Local, Remote (the words show as tooltips); the filter disappears as soon as one kind is gone.
  • Detached terminal window (the bottom bar moved to its own window): the remote tabs are listed there too, and clicking one opens it in that window, as a tab next to your local terminals, never over the agent of the main window. Clicking a local tab, or opening a new terminal, takes the screen back.
  • Backlog, a ticket in Todo: the small arrow next to Start opens Run on > This computer or one of your other machines that hold the project, each marked Online or Offline. A ticket running on another machine shows "On <device>" on its card. Since 2026-11-06 the same arrow sits next to the Start of each ready subtask in a card's Subtasks list (desktop and phone), so the subtasks of one group can run on different computers.
  • Phone: the Remote action of a project.
  • An agent that left a command or a sub-agent running over there reads In background rather than Done, with the same pill as a local agent (see Agent Background Work).

How to use it

  1. Sign in with the same account on each machine; they discover each other through the relay, nothing to configure on the router, no VPN, no port to open.
  2. Project on both machines: open the project here, pick an agent, Run on the other machine (or Add an agent on another machine: pick the Machine, then the Role on <device>). Its row shows "Runs on <device>"; opening it replaces the terminal pane while you keep your local files, changes and terminals. If that same agent also has a live session here, the local one wins: no remote badge, and clicking it opens the console on this computer.
  3. Project only there: click it in On my other machines ("This project is not on this machine: it opens as a remote workspace"). The remote view shows that machine's agents in a rail, grouped like the local list (Needs input, To review, Active, Seen, then Inactive below a separator); the rail scrolls and its add-agent button stays reachable at the bottom, its menu opening upwards (since 2026-09-23). If no project is selected, or the sidebar is folded into its rail, the same rail sits inside the view. To leave: Disconnect at the top right (the agents keep running there), any project of your sidebar including the one you came from, or Escape.
  4. An agent of your other computers has the same buttons as a local one (since 2026-11-07). Hover its row: the same panel comes out with Duplicate, Flash and Fork (the twin is created on that machine and opens here), the cross or, on an inactive row, Delete (the same confirmation as a local agent). The row's ⋯ button and its right click open the same menu: Open, Duplicate, Flash, Fork, Restart the agent, then the cross or Delete. The header of its pane has the same pencil next to the title (rename the session; the machine that runs the agent keeps the new title and every screen shows it), the same mail envelope with its unread count (the mail is read on that machine, without Reply or Extend, which stay on it), the same ⋮ menu (Files & Git, Restart the agent, Memorize now) and the same cross, or a bin for a quick agent. Restart the agent relaunches the CLI over there on the same conversation, and the console you are watching stays open and follows the new one. Since 2026-11-13 the model pill of the header opens the same Switch provider window as a local agent: pick the CLI, the model, the reasoning effort and how the conversation is carried over (summary, transcript, screen or new conversation), and the computer that runs the agent applies it exactly as if you had clicked there; the console you are watching follows the new CLI. There is no account choice from another computer (accounts are set up per machine, the agent keeps its own). The cross follows the local rule, applied by the machine that runs the agent: a quick agent, or the console of a ticket that is done, leaves the list with its session; any other agent stays in the list, ready to reopen. If the agent still runs something in the background, the same warning asks first. To leave without closing anything, click another agent. Not available from another computer: pane splits (one remote console at a time), and the ⋯ rows that read this computer's record of the agent (edit, pins, notifications, pause). An agent shared by another account keeps a written Stop on its row and in its header, and its cross only leaves the stream: copying or deleting it stays with its owner. A machine that is offline offers nothing but leaving. The host must run the same version: an older one simply stops the agent when asked to close it, ignores Delete, Restart the agent and the pencil, and shows no envelope. If its terminals are detached in their own window, Restart the agent does nothing, the limit agents_restart has too.
  5. Team run box: click a step or its console to attach to that console. A step with no running console stays clickable when its conversation can come back: a console asleep on that machine restarts on its conversation, a closed one reopens in a new console over there (When a Team Run Ends). The cross in the box header stops the run on its machine ("Stop this run on <device>", after the same "Cancel this team run?" confirmation as a local run; a finished run is dismissed instead). Hand-off and answers still belong to the host: "Hand-off and answers stay on <device>. From here you can open its consoles or stop it." A run stopped on its max cycles offers Add one cycle / Add two cycles here too: the buttons grey out while the request travels, and the band goes away once the other machine has resumed the run (a few seconds). A stalled step or a question the run is waiting on ("The team is waiting for you") is shown with the same banners as a local run, to be answered on the host.
  6. Remote terminal tab: click it to attach. What you type goes straight to that shell, so there is no composer and no confirmation. The badge at the top right names the tab and the machine; its detach icon stops following it ("Stop following this terminal. It stays open on <device>.") and its cross closes the terminal on the other machine ("Close this terminal on <device>"), the same way the cross of a local tab does. Your own local tabs stay in the bottom bar while you follow a remote session. Right-click the remote terminal for Copy (greyed without a selection) and Paste, then Clear buffer and Settings; there is no Restart, since the process lives on the other machine.
  7. Agent working on a backlog ticket on the other machine: its pane shows the same Ticket, Backlog and Done buttons as a local agent: at the top right of the terminal, at the start of the toolbar in the conversation view. Done marks the ticket done; the other machine then auto-commits, merges the worktree and closes the agent there, with the same rules as a ticket closed from the phone (it needs the project open on that machine). The buttons show when this computer's board is that project's and the ticket is in progress.
  8. Changes (N) in the header of a remote agent's pane lists the files that agent changed on the other machine (added, modified, deleted), its branch and the "Unpushed commits" count, read-only. It refreshes when you open it, each time the agent finishes a turn, and with its refresh button. "No uncommitted changes: everything the agent did is committed." means the work is committed. The Changes tab of the right panel shows the same list in its Code changes block while that agent's pane is open (titled "Changes on <device>"), as long as Follow the selected agent is on in the tab's view options; your local branch, pull and push stay above it, the agent pills and the commit box come back when you leave the remote pane. Since 2026-11-10, clicking a file in that Code changes block opens its diff, read on the other machine, in the same viewer as a local diff (← / → walk the files, Esc closes), marked "Read only, on <device>": no edit, rollback, stage or commit from here. A very long diff shows its beginning only, a binary file says so and asks you to open it on that machine. The Changes (N) popover of the pane header still lists the files only. Not offered for a project another account shared with you.
  9. Start a backlog ticket on another machine: in the board, click the small arrow next to Start on a Todo ticket and pick a machine under Run on. Start itself still runs the ticket on this computer, and the arrow only appears when another machine of your account holds the project. The other machine re-reads the ticket, refuses it if it already left Todo, then starts it exactly as if you had clicked Start there (its assigned agent or team, its worktree, its model). You get "Ticket started on <device>" or the reason it could not ("This project is not open on <device>.", "<device> found no agent to run this ticket...", "<device> did not answer..."). The card then moves to Development for every machine, shows "On <device>" with the agent's live status, and does not offer Resume here while it runs there.
  10. Move an agent with its conversation: right-click it > Continue the conversation on > a computer. The window says what happens ("<agent> picks up its conversation on <device>, where it left off. Its console on this computer closes.") and that only the conversation moves, not the files. Move packs the conversation, closes the console here, sends it, and the other computer resumes the agent in the background ("<agent> continues on <device>."). For an agent already running on another computer, This computer > Continue here brings it to the one you are using and opens it. If the move fails after the console here was closed, Resume here reopens the same conversation on this computer: nothing is deleted on the sending side.
  11. Quick agent on another machine: open the Quick agent lightning menu; at the bottom, Run on lists This computer and your other computers that hold the project (offline ones greyed with "Offline"). Picking a computer only selects it, the menu stays open; then click an agent tier or a model above and the quick agent starts on that computer, its console opening here like any remote agent. The provider, model and reasoning effort travel with it; the account is chosen by the other computer (accounts are set up per machine). The choice is for that launch only: the next time the menu opens, This computer is selected again.
  12. Conversation view of a remote agent (since 2026-10-20): the floating Conversation / Terminal button at the top right of its pane (the same button as a local pane) switches between the streamed terminal and the Conversation view, read on the other computer: the answers as a chat, the answer being written shown live, the question cards and the Allow / Deny (or trust this folder) cards clickable from here, the keys being typed on the machine that runs the agent. Which view opens first follows Settings > Terminal > Conversation view of the computer you are using. While the CLI over there holds a dialog no card can answer (a model picker...), the pane shows the terminal and comes back by itself. The composer and the header's buttons work the same in both views.
  13. When the other machine turns a request down you get the reason ("That machine does not have this project any more", "You do not have access to that agent on that machine") instead of a timeout.

Settings

  • deviceName (global, this computer only): Settings > Global settings > Machine name. The name this computer goes by everywhere in AgentsRoom: sidebar ("On <device>"), your other machines, the status bar list of machines, the mobile app, Chrome and shared rooms. Also editable from the status bar: click Machines, hover the "This computer" row and click the pencil next to its name (Enter saves, Escape cancels; other computers can't be renamed from there, each one names itself). Default empty = the system's computer name (hostname), shown as the field's placeholder. Saved when you leave the field or press Enter; your other machines and the phone pick it up within about 30 seconds. Only the label changes: the operating system's computer name stays the same and no admin rights are needed. Never synced: each computer names itself. Up to 64 characters.
  • openOnDeviceId (project, synced to your account, default empty): right-click a project in the sidebar (or its ⋯ menu on a project card) > Always open on <device>. Shown only while another computer of your account is online and has this project. Ticked: clicking the project (sidebar row, project logos, project cards, command palette) opens that computer's agents directly, as Open on <device> does, instead of this computer's workspace. When that computer is offline or no longer has the project, the click opens this computer's workspace as before. Pick the entry again to untick it. Default empty: a click always opens this computer's workspace, and Open on <device> stays the way to reach the other one. Agents set it with settings_set (scope project, key openOnDeviceId, the machine's id; empty to clear).
  • relayWsUrl (global, advanced): the relay the fleet rooms travel through. Leave the default.
  • No fleet-specific toggle to join: signing in on a second machine is the setup.
  • requireDeviceApproval (global, this computer only, default off): status bar Machines (or Mobile) popover > Approve new devices. The same switch covers phones and computers. On: another computer of your account (or a phone driving the fleet) must be approved on THIS computer, with a matching 8-character code on both screens, before it sees this computer's projects and agents, reads any output or does anything here. Approved devices are recognized by their own key and reconnect without asking. Turn it on on every computer you want to protect: each computer decides for itself. Computers seen while the switch was off are trusted when you turn it on; revoke any you do not recognize.

Agent tools (MCP)

None. projects_list and the agents_* tools act on the machine the agent runs on.

Providers

All providers, no difference. One machine can run Claude agents, another Codex, another Antigravity; they show side by side.

Moving an agent with its conversation: Claude Code and Codex agents resume their exact conversation on the other computer (the CLI's own conversation files move). Every other CLI starts a new session there that first reads the transcript of its console ("This CLI cannot reopen a conversation on another computer..."), which needs the console to be open when you move it.

Mobile

Yes: the Remote action of a project opens a machines > projects > agents screen and a remote terminal, without switching the paired desktop. Stopping works the same way there: since 2026-10-22 each agent row of the list has a Stop button (power icon) that stops it without opening its console, and the header of the remote console offers Stop ("Stop this agent on <device>?", "Its session ends there for good."). It follows the same rule as the desktop's cross (since 2026-11-07): a quick agent, or the console of a ticket that is done, leaves the list; any other agent stays. The back arrow leaves without stopping anything. A long press on an agent row of another computer opens its actions sheet (since 2026-11-07): Open terminal, Duplicate and Flash duplicate, with the same words as for an agent of the paired desktop. The twin is created on that computer, and its console opens on the phone. The project assistant has no sheet, and there is no Fork on the phone, as for the agents of the paired desktop. The pencil next to the title of the remote console renames the session on that computer, like the pencil of an agent of the paired desktop. Since 2026-11-13 its header also shows the agent's model pill, as for an agent of the paired desktop: tap it to open the same Switch provider sheet (provider, model, reasoning, how the conversation carries over) without the account, which stays the one of that computer. The sheet closes at once, the computer that runs the agent applies the change and the console you are watching follows the new CLI. A team run of another machine comes as a card inside its project (title, machine, counter, mode banner, steps) with its consoles listed under it; tapping a step opens a console, and the card's cross stops the run ("Cancel this team run?", "The run stops on <device> and its consoles close there."). A run stopped on its max cycles shows the same "This team used all its cycles" band as the paired desktop's run card, and Add one cycle / Add two cycles resume it on its machine. Hand-off and answers stay on the host, as on the desktop. The host's open terminals are listed after its runs and agents, one row per tab with a Running or Idle pill when the host reports activity and "<device> · terminal" under the name; tapping one opens the screen "Remote terminal on <device>", where the keyboard types into that terminal. The row's cross and the screen's Close button close the terminal on the host ("Close this terminal on <device>?", "<name> closes there, and whatever runs in it stops."). Every stop or close from the phone asks for confirmation first. The arrow next to Start on a Todo ticket also opens Run on on the phone. Choose the paired desktop or another online computer of your account that holds the project; offline computers stay listed and cannot be selected. The paired desktop relays the request without changing the pairing, and the selected host runs the saved ticket with its usual agent or team, model and worktree. Both the paired desktop and the destination must be online. An older paired desktop asks you to update it. The phone stays on the board and confirms the destination once the host answers. VoiceOver and TalkBack expose the machine choice through the card’s actions. The quick agent sheet (lightning button) ends with the same Run on rows: the paired desktop is selected by default, your other computers that hold the project are listed under it (offline ones cannot be picked). Select one, then tap a tier or Launch: the button shows "Connecting to the agent…" and the remote console opens once that computer answered, or the reason it could not is shown under the rows. The account choice is hidden while another computer is selected, since accounts are set up per machine. The remote console of an agent of another computer also has the conversation toggle at the top right (since 2026-10-20): Show the conversation replaces the output with the same chat view as an agent of the paired desktop (live answer, question cards, Allow / Deny), read straight on the computer that runs it, the keyboard staying below; Show the terminal brings the output back. Moving an agent with its conversation works from the phone too: the agent's actions sheet has a Continue the conversation on section ("Looking for your other computers…" while the paired desktop answers) listing your other online computers that hold the project, or the paired desktop's name ("Brings its conversation from <device>") when the agent runs elsewhere. A confirmation says "<agent> picks up its conversation on <target>, where it left off. Its console on <source> closes." with the same note about files; the paired desktop does the move exactly as if you had clicked there, and the phone shows "<agent> continues on <device>." or why it failed.

Limits

  • Moving an agent: both computers must be online, signed in to the same account, with the project open on the receiving one and the agent synced there (saved agents only: a ticket, team-run or quick agent cannot move). A project running on an SSH server cannot move its agents (its conversations live on the server). Files are not copied: commit and push the agent's changes, then pull on the other computer. A conversation larger than 256 MB compressed is refused. Both computers need a version from 2026-10-13 or later; an older one never answers ("did not answer in time").
  • Switching the model of an agent on another computer: that computer must have the chosen CLI installed (the provider list is the one of the computer you are on), and must run a version from 2026-11-13 or later (both computers need it); an older one ignores the request. The pill stays a label for an agent shared by another account, an agent that is not running there, or a computer that stopped answering. On the phone, the transcript handoff is not offered (as for an agent of the paired desktop), and the reasoning starts on Default: leave it untouched and the agent keeps its own.
  • Quick agent on another computer: that computer must have the chosen CLI installed (the provider list in the menu is the one of the computer you are on). Projects another account shared with you are never offered. A computer older than 2026-10-14 creates a plain custom agent on the chosen model instead of a quick agent.
  • Presence and control travel through the cloud relay even between two machines on the same LAN, so keystrokes take a round trip; there is no direct peer-to-peer path yet.
  • An offline machine keeps its place in the list, greyed out; its agents cannot be reached until it is back. Clicking the project still opens the remote view, which reads "<device> is offline" and offers Back; it shows the agents by itself once the machine answers again.
  • A custom uploaded avatar of an agent that exists only on the host does not travel; the row falls back to the role artwork.
  • Remote agents have their own terminals on the host; the browser, files and changes panels you see are those of the machine you are on. The Changes (N) button of a remote agent's pane, and the Code changes block of the Changes tab while that pane is open, are the exception: they read the other machine, read-only (diff of one file from the Changes tab, but no commit, push or rollback from here), and list at most 300 files.
  • The relay sees only encrypted blocks and an anonymous room id.
  • An agent started by a trigger on another machine is listed here only while its console is alive, marked Trigger, under the name of its trigger; a finished run stays in its trigger's Run history on that machine. Both machines need a version from 2026-11-03 or later: an older one announces nothing, and an older reader does not show these rows. A project another account shared with you never shows them. A team run a trigger started keeps its own box, like any other run.
  • An agent live on two machines at once is shown and opened locally; the remote copy is not offered a second row.
  • A remote team run box exists only while its machine is online and answering; the consoles of a silent machine stay listed, greyed, without their box. A step can take about ten seconds to change state on the other screen.
  • A project another account shared with you never shows its team runs: the box names the team, its steps and the subject of the run, which is not part of an attach grant. The host must run a version that announces its runs; an older one still lists plain agents.
  • Starting a ticket on another machine needs that machine online, signed in and holding the project; a project another account shared with you cannot start tickets on its owner's machines. The "On <device>" mark of a card comes from the live presence of the agent working on it: it disappears when that session ends or the machine goes silent.
  • Conversation view of a remote agent: both computers need a version from 2026-10-20 or later (an older host never answers: the view shows "Could not read the conversation." and the terminal still works). It shows the last 400 messages, without the previews of attached files (they live on the other computer) nor the session trace. Not offered for a project another account shared with you, nor for a terminal tab. The live answer needs the agent's pane mounted on the other computer, which attaching does.
  • Remote terminal tabs: only online machines are listed, and a tab of a machine that goes silent disappears with it. Database tabs are never listed. A project shared by another account never exposes its terminals (a terminal is a raw shell on the owner's machine, no access grant covers it). The tab cannot be dragged or split from here, and the host must run 1.182.0 or later (1.184.0 or later to close it or stop a run from afar). If the host closes the tab while you follow it, the pane keeps the last output received.

Common questions

  • How do I stop picking "Open on <device>" every time for a project I mostly run on another computer? Right-click the project (or its ⋯ menu) > Always open on <device> (reported by a user, 2026-11-16). From then on a click on the project opens that computer's agents while it is online; when it is off, the click opens this computer's workspace. The choice is saved with the project on your account, so your other computers follow it. Pick the entry again to go back to opening locally.

  • I launched an agent on my other computer and it stopped at once, or its header said "Custom agent · Default" with the Claude logo. Fixed on 2026-11-09 (reported by a user, both computers need this version). The other computer only read your saved agents when it opened the project, so an agent created since was unknown there: it now reads your account's agents again before refusing one. A quick agent created there shows the CLI and model you chose, and a launch that fails there ends the session with a message instead of a terminal that never fills. A throwaway agent is never synced, so it cannot be launched elsewhere.

  • Every agent of my other computer (or my Linux server) opens on "Session ended" and nothing else. That computer could not start the agent, most often because the agent's CLI (Claude Code, Codex...) is not installed there, or not signed in: an agent always runs with the CLI of the machine that runs it. Since 2026-11-16 (both computers need this version) the console says why before "Session ended": "<CLI> is not installed on <device>", "<device> has no room for another agent right now" (it then starts there by itself once an agent ends), or "The agent could not start on <device>". Install the CLI on that machine and sign in to it (on a server, from its terminal), then open the agent again. The same sentence shows above the message box (desktop and phone), so you see it in the conversation view too.

  • An agent on my other computer launched another agent: are they shown together here? Yes, since 2026-11-08 (the other computer needs this version too). In the agent list, an agent launched by another agent sits right under the agent that launched it, with the same branch line as for local agents, whichever computer runs them; the pair is filed in the section of whichever of the two needs you most. The phone's Remote screen nests them the same way. An agent launched by a computer that has not updated yet still shows as its own row.

  • I cloned a project on another computer and it shows up as a second project. Since 2026-11 a clone joins the project of your account when it opens, if it has no .agentsroom/project.json of its own: same remote, same project, with its name, agents and backlog, and a notice says so. Only a project with no folder on that computer can be joined: a second clone of a project already open there stays a separate row, and Duplicate always makes a new project. A clone that opened as a duplicate on an earlier version stays one on a desktop; on a server, agentsroom --server --open <folder> folds it back (see Linux Server Mode).

  • Can a new computer signed in to my account connect to my other computers without approval? Only while Approve new devices is off (the default). Turn it on in the status bar Machines popover on each computer to protect: a new computer then sees no project or agent of that computer and can do nothing there until you click Approve on it. The approval prompt and the new computer's Machines popover show the same code.

  • How do I revoke another computer? On the computer to protect, status bar Machines popover (with Approve new devices on), hover the computer, shield icon, Revoke. What it was driving here is disconnected at once (the agents keep running here), and it cannot come back until you approve it again. On the revoked computer, the row says so and offers Ask again.

  • My other computer says "Waiting for approval on …". That computer has Approve new devices on and has not approved this one yet: answer the prompt over there (check the code). After Deny or Revoke, click Ask again in this computer's Machines popover to send a new request.

  • Why did the Detach button disappear from a remote agent? It did the same thing as the cross of the pane. The cross now closes the agent like a local one, on the machine that runs it. To leave without closing anything, click another agent. A remote terminal tab keeps its detach icon.

  • Can I duplicate, fork or delete an agent of my other computer from here? Yes, from the panel that comes out of its row, like a local agent. The twin is created on that machine and opens here. Not for an agent shared by another account.

  • A trigger on my other computer started an agent and it waits for an answer. Can I answer from here? Yes: while its console is alive it is listed in the project with the mark Trigger and the name of its machine, it appears in the Waiting for input inbox, and clicking it opens its real console over there. What you type goes to that machine; the cross of its pane ends it there. Its link works the same way.

  • I clicked an agent of my other computer and, after 20 seconds, the app said that machine did not answer, although it was fine. Fixed on 2026-11-11: when it is THIS computer (or this phone) that has lost the relay, the request never leaves, and the app now says so at once: "This computer is not connected to the relay (network down or slow): the request never left. Check your connection, then try again." The message about the other machine is kept for the case where it really does not answer.

  • I switched to another project and came back: the agent of my other computer I was on is no longer selected, I get "What are we working on today?". Fixed on 2026-11-14 (reported by a user). Leaving a project still stops following that agent's console, but the project now remembers it: coming back opens it again, like a click on its row, after a short connection. Nothing is opened if the agent stopped over there in the meantime, or if you opened another agent of that project since.

  • Do I need SSH, a VPN or port forwarding? No. Both machines connect outwards to the relay; nothing is opened inbound.

  • Is my code synced between machines? No. Each machine keeps its own clone and runs its agents on it; only encrypted control messages and output travel, plus an agent's conversation when you move that agent to another computer.

  • Can I continue a Claude Code or Codex conversation on my other computer? Yes: right-click the agent > Continue the conversation on > the other computer. The conversation travels end-to-end encrypted through a relay room only the two computers join, the agent resumes it over there with its own resume flag, and the console here closes so the conversation is only written in one place. From the other computer, This computer does the same in the other direction for an agent that runs elsewhere.

  • I moved an agent but it does not see my latest changes on the other computer. Only the conversation moves. Commit and push on the computer where the agent worked, then pull on the other one.

  • How many machines? As many as you sign in on.

  • My computer shows up with an asset tag like "LT-4F2A9C". Can I give it a readable name? Yes: Settings > Global settings > Machine name, or the pencil next to "This computer" in the status bar Machines list, and type a nickname ("Work laptop"). AgentsRoom then uses it everywhere instead of the system name, on your other machines and on the phone too. It does not rename the computer in the operating system, so it needs no admin rights and works on a machine managed by your IT. Empty the field to go back to the system name.

  • I clicked a project under On my other machines and nothing loaded. How do I get back? Its machine is off or AgentsRoom is closed there. Since 2026-11-04 the view says "<device> is offline" and offers Back, which returns to the project or the home screen you came from. You can also click any project of your sidebar, or press Escape. Start AgentsRoom on that machine and the same view shows its agents, with nothing to reopen.

  • The other machine says "No answer". Check that AgentsRoom is open and signed in there ("Sign in on a second computer").

  • Can teammates join? Yes: invite them into a project and they attach to your agents with per-agent access. See Project Sharing.

  • A team run on my other computer has ended. Can I still see what its steps did? Yes, for the last three ended runs of the project that machine still shows, when their conversation can be reopened: click a step in the box, its conversation reopens in a new console on that machine. The ✕ removes the run there too.

  • I see the agents of a team run from my other computer, but not the team. Since the run box came in, the host announces its live runs and the agents regroup under them. If they still come as separate rows, the host runs an older version, the run is finished, or the project was shared to you by another account.

  • Can I hand off or cancel a remote run from here? You can stop it: the cross of the run box (desktop) or of the run card (phone) ends it on its machine after a confirmation, and its consoles close there. A run stopped on its max cycles can also be resumed from here with Add one cycle / Add two cycles. Handing off and answering an Await question still happen on the machine that runs it.

  • My team run on an unattended machine stopped on "Max cycles reached". Do I have to go to that machine? No. Since 2026-10-16 the stopped run shows in the project on your other computers and in the phone's remote view, with Add one cycle / Add two cycles (resume it over there) and the cross (dismiss it). Before, such a run was not announced at all, so nothing showed outside that machine and its paired phone. You can also ask the agent that orchestrates the run: see Agent Teams (Agent may add cycles).

  • Can I see the dev server running on my other Mac? Yes. Open the project here: its terminal bar lists the tabs open on the other machine (name, then the machine's name), as long as that machine is online. Click one to read its output live and type into it. To close it, use the cross of the tab or of the badge at the top right of the pane: the terminal closes on the other machine, exactly as if you had closed it there.

  • A remote agent's terminal is squeezed into a narrow column. A phone is driving it: the host sized it for the phone screen. The pane shows "The terminal has been resized for the mobile screen." with Take back control, which gives it your window's width again (since 2026-09-24, Grok agents included; the other machine needs a version that relays it, and your window may need a restart after an update).

  • I attached an image to a message for an agent on my other computer and it only received "File #1". Fixed on 2026-10-18: the file now lands on that computer before the message is typed, and the token is replaced by the file's path there, as with a local agent. Both computers need that version; an older host still types the path on its own. A file larger than 8 MB (a full-screen Retina screenshot can be) or that cannot be read cannot travel to another computer: since 2026-10-21 the message is not sent at all, it comes back in the input box with its files and a banner naming the file and the reason. Crop or compress the image, or copy the file to that computer and give the agent its path.

  • On my laptop I only see the raw terminal of the agent running on my desktop. Can I read it like a chat? Yes, since 2026-10-20: click the floating Conversation button at the top right of its pane. You get the same Conversation view as a local agent, with clickable question and permission cards, read from the other computer. Both computers must be up to date.

  • Can I copy text out of a remote terminal? Yes: select, then right-click > Copy, or the usual copy shortcut. Right-click > Paste types the clipboard into the remote shell.

  • Can I mark a ticket done when its agent runs on my other computer? Yes, from that agent's pane: Done closes the ticket for every machine, and the machine running the agent does the commit, merge and close there. Both machines need a version from 2026-10 or later.

  • How do I know whether the agent on my other computer committed its work? Open its pane and click Changes (N), or look at the Changes tab of the right panel: you get the list of uncommitted files and the unpushed commits on that machine, without asking the agent. Click a file of the Changes tab's list to read its diff. Committing and pushing stay on that machine.

  • Can I start a backlog ticket on my other computer without going to it? Yes: the arrow next to Start on a Todo ticket opens Run on, which lists your other machines that hold the project (online or offline; offline ones cannot be picked). The ticket runs there as if it had been started there, and its card shows "On <device>" with its live status on every machine. Both machines need a version from 2026-10-12 or later; an older one leaves the request unanswered ("did not answer").

  • Can I change the model of an agent that runs on my other computer, or on a headless server? Yes (since 2026-11-13): open its console and click the model pill in the header. The Switch provider window is the same as for a local agent, and the other computer applies the change, keeping the conversation the way you chose. The phone does the same from the header of the remote console.

  • Can I launch a quick agent on my other computer? Yes: in the Quick agent lightning menu, pick the computer under Run on at the bottom, then the tier or the model. The agent runs over there and its console opens here.

  • The remote terminal tab I saw a minute ago is gone. The other machine went offline, or the tab was closed there. A remote tab is a live session, not something kept for later like an agent.

  • A ticket started on my other computer: does my board show it live, and can I open its console? Yes. Within a few seconds of its start the card moves to In progress with "On <device>" and its live status, and when the run ends it moves on too. Clicking the card opens that agent's console from the other machine, as for a ticket running here (the eye button still opens the ticket); hovering "On <device>" says so. For a team ticket, the card follows the most urgent console of the run: the one waiting for your answer, otherwise the one working. Other edits made over there (title, description, column) still show at the board's next refresh, within 30 seconds. On the phone too (since 2026-11-10): while a project's board is on screen, the phone connects to your other computers, so a ticket started on one of them moves to In progress within seconds with "On <device>" on its card, and the card's Agent link opens that console on the phone (the screen of the Remote action). Tapping the card still opens the ticket, as for a ticket running on the paired desktop. Leaving the board closes that connection.

  • Two of my computers started the same ticket. Can it still happen? Not since 2026-11-06: the first start wins, and the other is refused with "This ticket has already been started." Nothing else is written, and the board of the second computer catches up.