Browser Use
used in
1
stack
Open-source framework that lets LLM agents control a real browser to complete web tasks.
Exposes browser actions to language models so agents can navigate, fill forms, extract data, and recover from errors autonomously. The 0.13 rewrite pairs a Python API with a Rust core and a browser harness designed for frontier models, and it supports multiple LLM providers plus custom tools, CLI usage, and real browser profiles for authenticated sessions. MIT licensed, with a separate Browser Use Cloud offering hosted stealth browsers, proxies, CAPTCHA solving, and parallel execution.
Strengths
- Agent-native loop with real browser actions, persistent tools, and error recovery
- LLM-agnostic, supporting Claude, GPT, Gemini, and a task-tuned ChatBrowserUse model
- MIT licensed with a large, active community
- Cloud option adds proxies, stealth browsers, and parallel execution for production
Tradeoffs
- LLM-driven navigation is slower and less deterministic than scripted automation
- Per-task token costs add up, and complex multi-step tasks can still fail or loop
- Fast-moving API with breaking changes between versions