browser-use/browser-use

Turns any website into something an LLM agent can actually navigate and act on, without requiring the site to have an API. You point it at a browser session, give the agent a task, and it clicks, fills forms, and extracts data like a human would. What separates this from the graveyard of browser automation wrappers: it is designed specifically for AI agent control flows, not for record-and-playback scripts. That means it handles the messy stuff, dynamic content, login walls, multi-step flows, without needing deterministic selectors that break the moment a product team ships a CSS rename. At nearly 94k stars it has clearly hit a nerve. Honest reservation: browser automation is brittle by nature, and LLM-driven automation is brittle squared. Plan for failure modes. But as a foundation for internal tooling or scraping pipelines, the abstraction level is exactly right. -> Best for: AI engineer building autonomous agents or solo founder replacing a fragile Playwright script with something that can reason its way through UI changes