Browser use

Browser use gives a sandbox a real, interactive browser. It is available on every sandbox, on any template, with nothing to choose at create time: the browser is part of the sandbox, not a separate machine. The live viewer is the one exception on SOMA today, noted below.

Inside the sandbox the browser runs on a virtual display (Xvfb :99), so it needs no desktop session. That display ships in the sandbox image. One exception today: a SOMA sandbox ships Docker and Chromium but not the virtual display the live viewer renders, so the live viewer is not available on SOMA yet.

A person can watch it, take control, or share a view-only link. An agent connects to the same browser over CDP, so automation and a human share one session, one set of cookies, and one page.

1. Create a sandbox

Any sandbox works, on any template.

Wait until the sandbox state is running.

2. Start the browser

The browser starts lazily on the first POST.

curl -X POST https://api.miosa.ai/api/v1/sandboxes/$SANDBOX_ID/browser \
  -H "Authorization: Bearer $MIOSA_API_KEY"

The response carries state, viewport, stream_url, cdp_url, and a short-lived stream_auth token. See Quickstart for every field.

3. Open it or drive it

A person opens the viewer, or an agent connects over CDP.

Stopping the browser with DELETE /browser frees its memory while the sandbox keeps running. The browser survives pause and resume.

Pick what you need

You want toStart here
Start and stop a browserQuickstart
Watch and take controlLive viewer
Drive it with an agentDrive it over CDP
Use a verified clientIntegrations
Check limits and billingLimits, billing, and security
Understand the modelHow browser use works

Next

Was this page helpful?