Browser Use

Browser Use is the open-source browser agent (Python and TypeScript) from the YC W25 startup of the same name ($17M seed led by Felicis, reported by TechCrunch March 2025; its approach gained wide attention after Manus built on it). Instead of driving the browser purely through screenshots, it converts page structure into a text-like format agents can reason over, then clicks, types, and navigates like a human would.

Three ways to use it: a Python library / TypeScript SDK you run yourself with your own model, a CLI/skill that plugs into Claude Code, Codex, Hermes or other agents, and a fully hosted Cloud offering (hosted agent + stealth browsers with CAPTCHA handling and residential proxies, billed per browser-hour) plus a hosted agent API.

1. Install with uv (Python ≥ 3.11): uv add browser-use.
2. Add your model API key to .env, e.g. OPENAI_API_KEY=... (optionally BROWSER_USE_API_KEY=... for cloud browsers / the BU2 model).
3. Write a small script (agent.py) that creates an Agent(task=..., llm=ChatOpenAI(model=...)) and calls await agent.run() — see the README quickstart for the exact snippet.
4. Run it with uv run agent.py; the agent opens a real browser and executes the task.
5. CLI path: paste the README's skill-install prompt into Claude Code/Codex/Hermes and run browser-use skill install to register it as an agent skill.
6. To disable telemetry, set ANONYMIZED_TELEMETRY=false in your environment (see the docs' telemetry page).