AI agents increasingly need a browser that can load pages, execute JavaScript and interact with web applications without carrying the resource overhead of a conventional desktop browser. Cloudflare (NYSE: NET) has open-sourced Kitesurf, its experimental stateless browser engine for AI agents, giving developers the option to inspect the code, run it locally or deploy an instance in their own Cloudflare Workers account.

The project is available on GitHub under the Apache 2.0 license. That changes Kitesurf from a managed capability into infrastructure that a team can modify and operate itself. For enterprises experimenting with agentic web workflows, the practical appeal is control over deployment and integration rather than dependence on a single hosted browser service.

Cloudflare introduced Kitesurf in August as a browser that runs entirely on Workers, then added WebMCP support, new APIs and compatibility improvements in September. The company says Kitesurf parses HTML, applies CSS, executes page JavaScript and renders screenshots without a browser container. It supports the Chrome DevTools Protocol, allowing clients such as Puppeteer to connect.

Cloudflare positions the design as a lighter alternative to running a full Chromium instance for every agent task. The company says Kitesurf is intended to load websites, execute JavaScript and interact with pages using less CPU and memory than a full browser while scaling across its global Workers network. That is a company claim, and the project remains under heavy development, so organizations should validate compatibility and resource behavior against their own sites and workflows.

The browser has three main parts: an Engine that manages sessions and client connections, PageScript that runs the page and its JavaScript, and PageRenderer that draws the result. Underneath, Kitesurf uses a collection of open-source Rust packages compiled to WebAssembly. Cloudflare identifies Blitz for HTML parsing, document structure, layout and painting; Stylo for CSS processing; Boa for selected JavaScript execution; and Linebender libraries plus Krilla for graphics and PDF output.

That architecture is relevant to IT teams evaluating an agent browser as a platform component rather than a black-box tool. The project separates its engine, Rust core, DOM shims, renderer and command-line interface into packages within the repository. Developers can build it locally, connect a Puppeteer client through a WebSocket endpoint, create a page, navigate to a site and capture a screenshot. The source also describes using a Chrome DevTools Protocol port for other agent harnesses.

Cloudflare says teams can deploy the engine to their own account after local testing, instead of using the Browser Run managed APIs. A Workers Paid plan is required because Kitesurf uses Dynamic Workers to create isolated Workers for pages. The company has also published build requirements, deployment guidance and contribution rules, while noting that it is still improving compatibility, reliability and efficiency.

Traditional browser automation frequently ties an organization to container fleets, full browser binaries and a managed service’s operational boundaries. Kitesurf offers a different model: an agent-oriented browser built on serverless primitives and now available for self-hosting. Its market impact will depend on compatibility, operational predictability and security controls, but the open-source release gives enterprise platform teams a more direct way to test whether that model fits their automation stack.