Hey HN, I built browser-use, an open-source alternative to OpenAI’s Operator for browser-use systems, and here’s why I think it’s better: Flexibility: You can use any LLM with our tool – Gemini, Anthropic, Qwen, Llama, DeepSeek, and more. As new models improve, so does your agent. Open Source: No need to pay $200/month or endure long waitlists – it’s free and accessible to everyone today. Custom Automation: Our Python package allows you to build actual web automations. Your LLM can gain new tools, like file uploads. Cost: Our system is 30x cheaper than Operator, e.g., when used with DeepSeek. Some models are even free, like Qwen or gemini-2.0-exp. Why it’s different from Operator: Usability: While Operator’s safety system is impressive, it often sacrifices usability. For example, their demo required logging in and email authentication – our tool connects seamlessly to your real browser. Full DOM Access: Unlike Operator’s vision-only approach, we process the entire DOM. If you want to extract hidden elements like links or large lists, we handle that directly in the input. Vision will never do that. Performance: We’ve already outperformed them on one benchmark, WebVoyager. Current benchmarks favor simple interactions, and we’re designed for much more. Who this is for: Developers who want to build automation or connect AI to the browser. Check out our browser-use GitHub repo. I would love feedback on how this fits into your workflows and what is needed for production readiness. Looking forward to your thoughts! |