skyvern-ai/skyvern

Skyvern automates tasks that normally require a human to sit in front of a web browser and click through a website. Think of jobs like filling out an insurance quote, downloading a statement from a banking portal, or stepping through a multi-page form. Traditionally, this kind of automation is written as a brittle script that looks for specific elements in a page's underlying structure; those scripts break the moment the website changes its layout. Skyvern instead uses large language models and computer vision to look at the page the way a person would, decide what to do next, and operate the browser through the Playwright automation library. Under the hood, Skyvern runs what it calls a swarm of agents that together read the page, plan the steps, and carry out the actions. Because the system reasons about what it sees rather than memorising fixed selectors, the README says it can work on sites it has never encountered before, keeps working when layouts change, and can apply a single workflow across many websites. On top of the engine, Skyvern ships an SDK that extends Playwright with AI-aware commands, for example asking the page to "click the login button" in natural language, extracting structured data against a schema, running multi-step tasks, and downloading files. There is also a no-code workflow builder for non-technical users, and a managed cloud version with anti-bot detection and CAPTCHA solvers. You would use Skyvern to automate repetitive browser work without writing fragile scrapers. It is a Python project, with a TypeScript client package, and can run locally via pip or Docker Compose, or via the hosted Skyvern Cloud.