FIELD GUIDE / BROWSER USE

Browser Use OSS and managed browser

Browser Use combines open-source browser-agent components with a separate hosted browser-and-agent offering.

Browser Use combines open-source browser-agent components with a separate hosted browser-and-agent offering. The distinction matters for security review: using a library locally does not automatically mean the team’s browsing data remains local once a cloud browser endpoint is configured.

How it works

The maintainer’s first-party repository describes a Python library and a hosted Cloud product with managed browsers and an agent task API. The maintained Cloud notes explicitly distinguish hosted-only components from related open-source projects. Treat those as distinct deployment types; do not apply an open-source license automatically to a hosted service’s terms or assume managed infrastructure is available in every geography.

A practical, bounded workflow

For a public information-gathering job, begin with a narrow list of allowed public sites and no stored user sessions. The agent locates an official document, extracts an identifier and URL, and sends the information through deterministic schema validation. Record the exact browser session, navigation origins, redirects and extraction confidence. An analyst then approves citations before the material is incorporated into a report.

Failure modes and limits

Hosted browsers can maintain cookies, profiles and proxy settings. These are part of the security boundary, not merely a speed feature. Attackers can insert instructions in pages that the agent reads; never permit page content to alter tool permissions, broaden visited domains or request secret extraction. Treat session state as sensitive even if the screenshots appear innocuous.

Deployment controls

Define each browser action as a proposal whose origin and destination are checked by the host application. Session creation, cookie storage, browser screenshots and download permissions must belong to a named principal and expire after the job. An agent-generated instruction may decide what to request, but the executor independently denies destinations and writes outside the approved ticket. Observe target state through a trusted read-only channel.

When to choose this approach

Use the library when control over browser placement and runtime matters and you can operate its dependencies. Use hosted browsing when managed browsers justify extra vendor and data-flow review. The page’s current pricing and limits are mutable, so all exact figures are intentionally omitted and must be sourced by AI 10.

What to evaluate before production

Test navigation to unapproved origins, misleading banners, a logout, a blocked CAPTCHA, changing layouts and stale sessions. Check that the agent fails safely and records partial results without presenting an incomplete extraction as a successful completed task.

Version, market and tier caution (25 October 2026): This is a technical and operational guide, not a live entitlement, pricing or geographic-availability certification. Numeric commercial limits and undocumented release dates are intentionally omitted. Reconfirm the exact account, deployment region, access policy and vendor release before procurement or publication.

Sources and Further Reading

CONTINUE EXPLORINGExplore related files ↗