Computer use vs browser use: which to pick
Both drive a real browser. They differ in how much machine you rent.
| Browser use | Computer use | |
|---|---|---|
| What runs | A live browser inside a sandbox | A full Linux desktop on a Computer |
| Where | Any sandbox, any template | A Computer you create and start |
| Setup | On by default; nothing to choose at create time | Create a Computer, then start it |
| Live viewer | On the Sandbox page: take control, view-only share links, full screen | A streamable desktop, embeddable with a short-lived URL |
| Agent access | Over CDP - Playwright, Puppeteer | Screenshot and control APIs: mouse, keyboard, windows, accessibility |
| Human takeover | The browser view | The whole desktop |
| Other apps | No - browser only | Yes - any installed Linux app |
| Persistence | Survives pause and resume | Survives stop and start |
| Billing | Plain sandbox usage | Computer compute while running |
| Best for | Web automation, scraping, logged-in browser work that should be watched | GUI apps, desktop workflows, screenshot QA, full human handoff |
Pick browser use when
- The task lives entirely on web pages.
- You want the cheapest isolated environment: a sandbox you already use for code.
- A person should be able to watch, take control, or share a view-only link.
- Playwright or Puppeteer is your driver.
Pick computer use when
- The task needs desktop apps, file pickers, extensions, or multiple windows.
- The agent works from pixels and the accessibility tree across the whole screen.
- A human must take over the exact machine an agent was using.
- You need a durable desktop that keeps its browser profile and installed tools across restarts.
Next
- Browser use - start a live browser in any sandbox.
- Computer use - the desktop perception-action loop.