Skip to content

Install Browser Use 0.13.10 for LLM Automation

Deploy Browser Use 0.13.10 for LLM browser automation: system specs, uv install, Chromium setup, doctor checks, CLI 3.0 traps, and cleanup.

5 min readIntermediate

Here’s the bit most LLM browser automation posts still get wrong: starting with Browser Use 0.13.x, the old CLI subcommands (open, click, state) are gone. CLI 3.0 hands agents a Python eval loop on top of Browser use. Copy a 2025 install guide and nothing responds? That’s why.

This is a deploy guide for Browser Use 0.13.10 (released 4 Sep 2026 per the GitHub latest release) – the open-source library that lets an LLM drive a real browser. Pin the version, install Chromium on purpose, run doctor, stop.

System requirements before you touch pip

Official installation docs set this floor. Don’t under-spec the box if a headed Chromium stays open for long agent runs.

Resource Minimum Recommended
Python 3.11+ 3.12 (current quickstart)
OS macOS, Linux, Windows (incl. WSL) native macOS/Linux for local Chrome attach
RAM 4 GB 8 GB+
Disk ~500 MB (package + Chromium) 1 GB+ free for caches/profiles

requires-python in package metadata is >=3.11,<4.0 (see the repo pyproject.toml). No matching distribution found for browser-use almost always means the active interpreter sits outside that range – or pip is talking to the wrong Python.

Official download source (don’t freestyle this)

Release notes for 0.13.10 bump Browser use to 0.1.13, pin Pydantic 2.13.5, move the MCP Python SDK to 2.1.1, and add pydantic-settings 2.15.0. Old lockfiles still resolving 0.12.x need a forced upgrade.

Install Browser Use 0.13.10 step by step

Library path first (code-first agents). CLI alternate at the end.

  1. Install uv if needed, then a 3.12 venv:
pip install uv
uv venv --python 3.12
source .venv/bin/activate
# Windows: .venvScriptsactivate
  1. Pin the library:
uv pip install 'browser-use==0.13.10'
# or in a uv project:
# uv add 'browser-use==0.13.10'
  1. Install Chromium for local runs. The wheel alone is not enough – this is the half-broken setup trap:
uvx browser-use install
# equivalent: python -m browser_use install
# under the hood: uvx playwright install chromium
# (Linux adds --with-deps)

Pip route: python -m venv .venv, activate, pip install 'browser-use==0.13.10', then python -m browser_use install.

CLI-only (coding-agent workflows), from the CLI docs:

uv tool install --python 3.12 --upgrade --force browser-use
browser-use --help
browser-use --doctor

Docker is path three: clone the repo and docker build -t browseruse . (or Dockerfile.fast after base images). Containers auto-disable the Chromium sandbox – leave that alone.

First-time configuration (minimum that runs)

You need an LLM key. The open-source package is free to run locally; docs push ChatBrowserUse with a Browser Use Cloud key. Eligible new Google/GitHub/Microsoft cloud signups get a one-time $15 credit as of the Sep 2026 README (this may have changed since).

# .env
BROWSER_USE_API_KEY=your-key-here
# or bring your own:
# OPENAI_API_KEY=...
# ANTHROPIC_API_KEY=...
# GOOGLE_API_KEY=...

Smoke agent:

from browser_use import Agent, ChatBrowserUse
from dotenv import load_dotenv
import asyncio

load_dotenv()

async def main():
 agent = Agent(
 task="Open example.com and return the page title",
 llm=ChatBrowserUse(),
 )
 history = await agent.run()
 print(history.final_result())

asyncio.run(main())

Save as smoke.py. Run uv run smoke.py or plain python smoke.py inside the venv.

Verify the install actually works

Green pip means nothing by itself. Three checks:

python -c "import browser_use; print(browser_use.__version__)"
# expect: 0.13.10

browser-use --doctor
# checks Python, browser binary, API key presence, connectivity

CLI 3.0 sanity on Unix:

browser-use <<'PY'
new_tab("https://example.com")
print(page_info())
PY

PowerShell wants a here-string piped in, not <<'PY'. The catch is attach: doctor green, browser still dead? Chrome is waiting for you to approve remote debugging – allow it, rerun.

Common install errors and fixes

GitHub issues and chat threads – not the polite FAQ.

  • No matching distribution found for browser-use – wrong Python. Recreate the venv on 3.11-3.13 inside >=3.11,<4.0, reinstall. Skip the system interpreter from a random IDE terminal.
  • Package imports, browser never opens – skipped uvx browser-use install. Run it. Linux may still need Playwright system deps.
  • browser-use: command not found on zsh – older curl|bash installers wrote PATH into ~/.bashrc while you live in zsh (see community threads around issue #4440 / PR #4457). Fix PATH in ~/.zshrc, or uv tool install --python 3.12 --upgrade --force browser-use so the shim lands correctly.
  • Old scripts calling browser-use open ... – subcommands died with CLI 3.0 (0.13.3+). Heredoc Python helpers, or drive the Python Agent API only.

I still remember the first time I “finished” install, ran an agent, and got a perfect Python stack with zero pixels – Chromium was never on disk. That gap between library installed and browser installed is the whole game.

Upgrade from older versions / uninstall

From 0.12.x, library Agent(...) calls are mostly unchanged. Pin hard so uv does not leave a stale 0.12 lock:

uv pip install -U 'browser-use==0.13.10'
# uv project:
uv lock --upgrade-package browser-use
uv sync

uvx browser-use install

Full cleanup:

uv remove browser-use
# or: pip uninstall browser-use

uv tool uninstall browser-use

rm -rf ~/.browser-use
# Windows: rmdir /s %USERPROFILE%.browser-use

What still isn’t spelled out cleanly on one official page: whether every optional extra wheel is required for a basic local Agent run on your exact CPU/OS combo. If doctor names a missing binary after a plain install, add that extra – don’t guess from blog posts.

FAQ

Do I need Browser Use Cloud to run locally?

No. Local Chromium (bundled or system) is enough. Cloud is optional – stealth browsers, proxies, hosted agents, and Browser Use API models if you pick them.

uv tool install vs uv pip install – which one?

Importing Agent in a project? uv pip install / uv add inside that venv. Want a global browser-use binary for coding agents? uv tool install browser-use. Plenty of setups keep both: library in the project, CLI as a tool.

I upgraded to 0.13 and my shell scripts died. What changed?

CLI 3.0 killed fixed subcommands. Send use Python on stdin (or stop using the CLI and call Agent from Python). Run browser-use --doctor, then rewrite the wrappers.

Next: create the 3.12 venv, run uv pip install 'browser-use==0.13.10' && uvx browser-use install, then browser-use --doctor before any task prompt.