Why give an AI agent its own computer?

An AI agent with its own computer works apart from your devices, keeps going when you close your laptop, keeps its files and sign-ins between tasks, and can be rolled back when it makes a mistake.

Updated

Three places an agent can work

Your own computerA browser sessionA computer of its own
SetupInstall an appNoneNone, or self-host
Can reach your filesYesNoOnly what you share
Works while you're awayOnly if your machine stays onFor a short sessionYes
Keeps files and sign-insYesUsually notYes
If it makes a mistakeIt happened on your machineLimited to the sessionLimited to its machine; restore a checkpoint
Can run code and install toolsYesNoYes

What an own computer makes possible

The costs

One computer for all agents, or one each?

Some products give every agent its own machine. Others run all of a person's agents on one shared machine. One each means a research agent that installs odd software can't break the assistant that handles your mail, and each can be limited to the sites it needs.

Two diagrams side by side. On the left, three bots all feed into one shared computer that holds one set of files and a shared setup, so what one bot does can affect the others. On the right, three bots each have their own computer with their own files and sign-ins, so each bot is walled off from the others, as in Casama.
Bots on one shared computer share its files and setup. In Casama, each bot works on its own computer.

How Casama does it

Every Casama bot works on its own isolated Linux machine with a shell, files and a browser, and optionally a full desktop. Computers sleep when idle and wake with everything intact. You can watch one live, take control, and restore a checkpoint taken before a destructive command. Logins from 1Password or Bitwarden are filled in on the way out and never shown to the model. If you'd rather a bot used your own machine for a task, you can lend it through the desktop app, where high-risk actions always ask.

Join the waitlist Run it on your own server