Cloudflare Makes Kitesurf Open Source for AI Agent Browsing

Cloudflare has released Kitesurf under Apache 2.0, letting developers modify its browser for AI agents and run their own instance. Deploying it to a Cloudflare account requires a Workers Paid plan.

Nov 14, 2026
•
4 min read
Technobezz
Cloudflare Makes Kitesurf Open Source for AI Agent Browsing

Don't Miss the Good Stuff

Get tech news that matters delivered to your inbox.

Cloudflare has made Kitesurf, its browser for AI agents, open source under Apache 2.0, giving developers access to code they can modify, run locally and deploy to Cloudflare Workers. The code is on GitHub, and developers can operate an instance through their own Cloudflare account instead of using the company's managed Browser Run APIs. Deployment to an account requires a Workers Paid plan because Kitesurf relies on Dynamic Workers to isolate individual pages.

Kitesurf handles HTML and CSS, runs JavaScript from websites and produces screenshots, with all of that processing taking place on Workers without a browser container. It accepts connections through the Chrome DevTools Protocol (CDP), allowing developers to use clients including Puppeteer. Cloudflare says its aim is to let agents navigate and interact with websites while consuming less processor time and memory than a full browser such as Chromium, with capacity scaling across its Workers network.

Read more: Cloudflare Adds Personal Agent Analytics and Access Controls

The open source release comes with a development caveat: Cloudflare says Kitesurf remains under heavy development, with compatibility, reliability and efficiency among its priorities. Cloudflare introduced the browser in August, then added WebMCP support, further APIs and compatibility improvements in September. Both earlier updates included a commitment to release the code so developers could operate the engine through their own accounts.

For local use, developers must meet the repository's build requirements, clone the code and install dependencies before starting the engine. The initial build compiles a Rust core and its dependencies; later builds can draw on cached results. Wrangler then starts a local development server that developers can access through Chrome DevTools or connect to programmatically, including through an AI harness using CDP.

The architecture divides responsibilities among Engine, which handles sessions and connections, PageScript, which executes pages and their JavaScript, and PageRenderer, which draws the output. Its dependencies include Blitz for page structure and layout, Stylo for CSS processing and Krilla for generating PDFs. Those projects keep their own licenses, with third-party notices recorded in the repository's NOTICE file.

Cloudflare is accepting contributions, but asks prospective contributors to obtain a maintainer's endorsement before submitting a pull request. Developers can also contribute examples that reproduce rendering problems, automation failures or documentation errors, alongside narrowly scoped code fixes supported by tests.

Share