Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex and Gemini CLI read it natively; Cursor and VS Code connect over MCP.

The voice runs on your own machine, so there is nothing to meter. The first run downloads it, about 340 MB.

Yes. The skill is MIT-licensed and the authoring format is public.

Create

Browser Use 0.13.14 finishes OpenAI's browser tools and adds Bash

54 minutes ago

@browser-use Subscribe

Browser Use 0.13.14 completes the OpenAI browser tools: all 28 browser tools plus Bash, with scoped page reads, console and network diagnostics, and 10 MB uploads and downloads. Also: structured, bounded page reads, a single clear error for a missing Browser Use Cloud key, and reconnect cleanup. The last release on the Python agent before the Pi migration. Release notes: https://github.com/browser-use/browser-use/releases/tag/0.13.14 OpenAI browser tools guide: https://docs.browser-use.com/open-source/customize/integrations/openai Full changelog: https://github.com/browser-use/browser-use/compare/0.13.13...0.13.14

Ask about this presentation

Show transcript

Browser Use 0.13.14 finishes OpenAI's browser tools. The driver now exposes all twenty eight browser tools, adding scoped page reads, console and network diagnostics, uploads and downloads, plus a Bash tool.
Slide text
+ 01
Browser Use 0.13.14 finishes OpenAI's browser tools and adds Bash
All 28 browser tools and a Bash tool for OpenAI's Responses tool runner
Release 0.13.14 · PR #6092 · browser-use/browser-use
Uploads and downloads run through single-use grants. Each grant covers one file on one page, files stop at ten megabytes, and downloads reject redirects. Larger Cloud files still have to wait.
Slide text
+ 02
Uploads and downloads, up to 10 MB a file
+ Each grant covers one file on one page, once
+ Downloads reject redirects
+ Larger Cloud files wait for native Cloud upload and download
PR #6092 · browser-use/browser-use
Page reads now keep the page's hierarchy, field values, and which boxes are checked. Output stops at ten thousand characters, and when reading fails, it retries once without ever replaying a click.
Slide text
+ 03
Page reads keep the page's structure
+ Hierarchy, field values, checked and selected states
+ Reads, page text and scripts capped at 10,000 characters
+ A failed read retries once; clicks and typing never replay
PR #6093 · PR #6099 · PR #6092
Start a Browser Use Cloud browser without an API key, and you now get one error that says where to get a key. It replaces a hundred and thirty three log lines and three tracebacks.
Slide text
+ 04
A missing Cloud key now prints one error
ERROR [Agent] Agent run failed: BROWSER_USE_API_KEY is not set. 1. Create a key (free to start) 2. Set it: export BROWSER_USE_API_KEY=…
PR #6112 · browser-use/browser-use
Three smaller fixes. A failed reconnect now closes its half-built connection, the Cloud SDK's version four import works, and Cloud browsers stop telling you to try Cloud at a captcha.
Slide text
+ 05
+ A failed reconnect no longer leaves a half-open browser connection
+ Browser Use Cloud SDK V4 imports now install alongside Browser Use
+ Cloud browsers stop suggesting Cloud when they hit a captcha
PR #6102 · PR #6098 · PR #6112
This is the last release on the Python agent before the Pi migration. PyPI still serves 0.13.11 until a second release reviewer approves. Driving Browser Use with OpenAI models? This is the version I would install.
Slide text
+ 06
The last release on the Python agent before the Pi migration
On PyPI after a second reviewer approves
Release notes: Browser Use 0.13.14
github.com/browser-use/browser-use/releases
Release 0.13.14 · PR #6220