All stories
Browser Use Ecosystem Community reports

Browser Use gives OpenAI agents a local or cloud browser

Version 0.13.13 connects the OpenAI SDK's browser tools to Browser Use Actor, with 18 default tools and six optional actions. The release also fixes cloud tab URLs and scaled-display coordinates.

By AITrending2 min read
Sources (5)

Why now Browser Use published 0.13.13 on October 15, 2026 at 23:50:45 UTC. The dated release establishes the integration event; a repository push does not. Reporting is pinned to this tag because the live guide already describes later features.

Browser Use 0.13.13 lets applications connect the OpenAI SDK’s browser tools to a local browser or Browser Use Cloud. Released late on October 15, the integration uses Browser Use Actor to perform actions while OpenAI’s SDK handles the model loop.

The tagged implementation exposes 18 default browser tools and six optional actions. Applications can create an owned local or cloud session, or supply an already connected BrowserSession. The driver closes sessions it owns; a borrowed browser remains the caller’s responsibility. This is a Python integration, rather than a new model or a Codex plugin.

A concrete path for agent developers

The release’s quickstart imports BrowserUse from browser_use.integrations.openai, then connects it to AsyncBrowserToolset and AsyncResponsesToolRunner. It demonstrates a deliberately narrow task: open example.com and read its text, with three exposed operations, a URL policy restricted to that HTTPS host and a six-step limit.

The package requires Python 3.11 or newer, below 4, plus the openai-tools extra, which depends on OpenAI Python 3.31 or newer, below 4. A source-pinned installation route for the reviewed release is:

uv add "browser-use[openai-tools] @ git+https://github.com/browser-use/browser-use.git@129853546ff9cc169977d38b6038f9c4a4e734db"

AITrending’s version-specific PyPI lookup returned 404 during review, so package-index availability for 0.13.13 was not confirmed. The command above targets the published source commit. Local use also needs a browser installation; the project documents uvx browser-use install for Chromium. The example needs an OpenAI API key and OPENAI_MODEL; its cloud mode additionally requires Browser Use credentials. MIT-licensed code does not include model API or hosted-browser costs.

Fixes, permissions and the release boundary

The release maps a temporarily empty cloud-tab URL to about:blank and corrects screenshot, crop and pointer alignment on scaled displays. These changes address concrete failures in browser-driven tasks. The implementation also binds page element references to a tab and invalidates them after navigation or a new page read.

An authenticated browser may already have access to accounts and data. Applications therefore need to choose exposed actions, URL policy and confirmation behavior for their workflow. The quickstart demonstrates those controls; it does not establish unrestricted browsing as a safe default.

The integration pull request reports a mixed five-task live pilot: CDC and PlayStation tasks passed, while Southwest, EPA and Good Food tasks stopped. These are the project’s observations, not an AITrending execution test or a general reliability score.

The live documentation has already moved beyond this tag to a larger tool set and Bash support. Those later capabilities are outside this 0.13.13 report. The immediate news is an inspectable adapter that lets OpenAI SDK applications use Browser Use’s existing browser infrastructure.

Related reading