Skip to content
INTERNET INFRASTRUCTURE & NETWORKS

Cloudflare Open Sources Kitesurf, Its Stateless Browser Engineered Specifically for AI Agents

Cloudflare has officially fulfilled a promise made earlier this year by open-sourcing Kitesurf, its innovative stateless browser designed from the ground up for autonomous AI agents. Published under the permissive Apache 2.0 license on GitHub, the release allows developers worldwide to inspect the source code, run custom instances locally, modify the underlying architecture, and deploy their own browser engines directly onto Cloudflare Workers.

The unveiling marks a significant milestone for the project, which was initially introduced in August alongside subsequent updates in September that brought WebMCP support, new application programming interfaces, and enhanced compatibility. From its inception, Cloudflare signaled its intention to give developers the freedom to run their own browser infrastructure on their own accounts, a vision that is now fully realized.

Unlike traditional web browsers that rely on heavy resource containers, Kitesurf operates entirely on top of Cloudflare Workers. It handles the parsing of HTML, the application of stylesheets, the execution of complex page JavaScript, and the rendering of visual screenshots without ever spinning up a traditional browser container. By bypassing the heavy lifting typically associated with standard engines like Chromium, Kitesurf achieves a dramatic reduction in CPU and memory consumption. Furthermore, it leverages Cloudflare’s expansive global network to scale effortlessly on demand, providing a uniquely optimized environment for automated workloads.

A central feature of Kitesurf is its native support for the Chrome DevTools Protocol, commonly known as CDP. This compatibility layer allows standard automation clients, such as Puppeteer, to connect seamlessly to the browser engine. Developers and AI engineers can point their existing automation harnesses directly to the CDP port, granting autonomous agents full, granular control over page navigation, interaction, and data extraction.

Beneath the hood, the Kitesurf repository brings together Cloudflare’s robust developer platform APIs and an advanced ecosystem of open-source Rust packages compiled efficiently to WebAssembly. The architecture is divided into three primary components: the Engine, which expertly manages active browser sessions and incoming client connections; PageScript, which handles the execution environment for the page and its embedded JavaScript; and PageRenderer, which draws the final visual output. These modular parts, alongside a dedicated command-line interface, reside within the repository’s structured directory, offering deep visibility into how the disparate technologies integrate.

Kitesurf is now open source

For developers eager to test the technology locally, the repository provides straightforward paths for local builds and execution. Setting up an instance involves cloning the source code, installing dependencies, and launching the development server via Wrangler. Because the initial build process compiles the Rust core and its associated dependencies, the first-time setup requires a brief waiting period before the local development server goes live at localhost. Subsequent builds benefit from cached assets, significantly speeding up the initialization cycle.

Once running locally, developers can interact with Kitesurf using standard Node.js scripts via Puppeteer or direct CDP connections. For instance, connecting a Puppeteer script to the local WebSocket endpoint allows automated instances to open new pages, navigate to target URLs, and capture high-resolution screenshots with minimal configuration overhead. AI harness frameworks can similarly hook into the local CDP port, empowering automated agents to execute complex web interactions swiftly and reliably.

Beyond local experimentation, developers who have completed their testing cycle can transition their custom instances directly to their own Cloudflare accounts, shifting away from the managed Browser Run APIs provided by the company. Because Kitesurf utilizes Dynamic Workers to isolate individual pages securely, deploying the engine to a personal account requires a Workers Paid plan.

As the project enters the open-source ecosystem, the development team is welcoming forks, discussions, and code contributions from the broader community, though development remains under heavy active iteration. Current engineering priorities are squarely focused on enhancing the engine’s core compatibility, as tracked through Web Platform Tests, alongside continuous improvements in reliability and computational efficiency.

Contributors looking to participate are encouraged to consult the repository’s established contribution guidelines, which recommend opening preliminary issues with minimal reproductions or proposed design approaches before submitting pull requests. Beyond direct code contributions, the project welcomes bug reports, failing automation workflows, and documentation refinements to help propel the platform forward. As Kitesurf continues to evolve at a rapid pace, the team encourages developers to follow ongoing updates through the official company blog, developer changelogs, and community channels on Discord as they refine the future of browser automation for artificial intelligence.