stagehand

Why we included this project

Browser automations tend to break the moment a site changes its layout, and Stagehand is aimed at that exact problem. It sits between low-level scripting in Selenium, Playwright, or Puppeteer and the unpredictable black-box agents that can't be trusted in production. You write the steps you know in code, and hand the fuzzy parts, like clicking the right button on an unfamiliar page or pulling structured data out of it, to an LLM through simple act(), extract(), and agent() calls. Self-healing and action caching mean repeatable steps run without LLM inference until the page actually changes, which keeps both cost and flakiness down. It ships as a TypeScript SDK with a separate Python implementation.