Docs
Desktops & live view
The agent sees a screen and clicks and types on it. On a Mac that is the Mac's own screen; on Linux every conversation gets its own virtual desktop.
The live view
- Watch: Mob shows the screen the agent is using, live. On Linux that is the open conversation's desktop.
- Take control: switch from View to Control to use the mouse and keyboard yourself, for a sign-in or a captcha, then switch back.
- Quality: Auto, High or Low per machine in Mob.
- Transport: WebRTC over UDP (port 9527), or a WebSocket where UDP can't pass, such as through a Cloudflare Tunnel.
On macOS
- One screen for everything: the agent works on the Mac's own screen, with the apps and logins of the logged-in user.
- One turn at a time: because conversations share that screen, their turns wait in a queue; Mob shows the queue.
- Permissions: Screen Recording and Accessibility (how).
- Awake: the Mac must not sleep; the display may, the live view wakes it.
On Linux
Each conversation gets its own virtual desktop: an X display with a window manager, invisible on the computer itself. Conversations don't share a screen, so their turns run side by side (up to maxRunning, 4 by default).
| When | What happens |
|---|---|
| The conversation's first turn | its desktop starts |
No turn, viewer, CPU use or screen change for desktopSleepAfter (10 minutes) | it sleeps: frozen, using no CPU, windows kept |
| The next turn, or opening the live view | it wakes as it was |
Archiving or deleting the conversation, Close desktop in Mob, or mobd desktops close | it closes; the next turn starts a fresh one, with the conversation's files and browser profile kept |
mobd stop | asks whether to close desktops; kept ones re-attach on the next start |
Mob shows each conversation's desktop state, and a banner when memory or disk runs low, listing the desktops you can close. On the office computer, mobd desktops list shows them and mobd desktops logs SESSION their logs.
Browsers and sign-ins
On Linux, Chrome, Chromium and Firefox started by the agent use a browser profile and Downloads folder of the conversation's own, so conversations don't share cookies or sign-ins.
To start new conversations already signed in: in one conversation, sign in to the sites (from the live view, in Control), close the browser, then make its profiles the template:
mobd desktops list # find the conversation's ID mobd desktops seed SESSION # its profiles become the start of every new conversation's
On macOS the agent uses your normal browser with your own profile.
Limits on Linux
- Memory: every open desktop has its own apps. Below
memoryWarnPercent(15%) available memory Mob warns; belowmemoryFloorPercent(5%), or with too little disk, no new desktop starts; running ones keep going. mobd never closes a desktop to free memory: close those you don't need. - No GPU: pages render in software. Normal sites work; WebGL-heavy sites and video are slow.
- No sound.
- One size: every desktop is
desktopSize(1920×1080), also when watched from a phone; Mob scales the picture. - Not the computer's own screen: a Linux PC's physical display is not watched or controlled.
- Not isolated: desktops, agents and mobd run as one user; agents can read each other's files and profiles.