You can now self-host Kitesurf, Cloudflare's open-source browser for AI agents
Cloudflare has released Kitesurf, an open-source browser for AI agents, allowing developers to self-host the tool and inspect, modify, and deploy it on their systems.

Cloudflare has unveiled Kitesurf, an open-source browser designed specifically for AI agents, enabling developers to self-host the application. This move allows users to load web pages and execute JavaScript on Cloudflare Workers without needing a full desktop browser. Developers can now access Kitesurf's source code, giving them the ability to inspect, modify, and deploy the tool on their own machines or Cloudflare accounts, enhancing customization and integration capabilities.
Kitesurf is engineered to operate without a browser container, offering the ability to return screenshots of loaded pages. It supports the Chrome DevTools Protocol (CDP), which is commonly used by automation tools to control browsers. This compatibility allows clients like Puppeteer to connect seamlessly. Cloudflare's aim with Kitesurf is to provide a lightweight alternative to Chromium, using less CPU and memory, while also being scalable across its Workers network, which is a platform for running serverless applications.
The functionality of Kitesurf is divided into three main components: the Engine, PageScript, and PageRenderer. The Engine manages sessions and connections; PageScript executes the web pages and their scripts; and PageRenderer converts the output into images. Kitesurf relies on several open-source Rust packages compiled to WebAssembly, including Blitz for HTML parsing, Stylo for CSS, Linebender libraries for text layout and graphics, and Krilla for PDF generation. JavaScript execution is handled by Boa within WebAssembly, used only when necessary.
Setting up a local copy of Kitesurf requires meeting certain build requirements and executing four commands: cloning the repository, navigating into it, installing dependencies with pnpm, and running pnpm run dev. The initial build involves compiling the Rust core and its dependencies, which can take some time, though subsequent builds benefit from cached data. Deployment is achieved with the command pnpm run deploy, requiring a Workers Paid plan due to the use of Dynamic Workers for creating isolated environments.
- The specific impact on performance metrics compared to existing solutions is not detailed.
- The timeline for future updates or additional features is not mentioned.
- Potential security implications or vulnerabilities have not been discussed.