Cloudflare Open-Sources Kitesurf, a Browser for AI Agents That Runs Without Chromium
Cloudflare has open-sourced Kitesurf, a browser for AI agents that needs neither Chromium (the engine behind Chrome) nor a container for each session. It runs on Workers, Cloudflare's cloud platform, parses HTML, CSS, and JavaScript on its own, and takes screenshots. It connects like a regular browser via Puppeteer over the CDP protocol, and the Apache 2.0 license allows modifying the code and deploying it yourself. For teams running many agents on the web, this is a chance to move away from a heavy fleet of Chromium instances.
AI-processed from Cloudflare AI Blog; edited by Hamidun News
Imagine your AI agent needs to visit a website, wait for the scripts to load, click a button, and take a screenshot. Typically, this means spinning up Chromium, the engine that powers Chrome, inside a separate container for every session. It eats CPU and memory, starts slowly, and scales poorly when you have not one agent but thousands.
On November 13, 2026, Cloudflare open-sourced Kitesurf, a browser that tackles this problem. It runs entirely on Workers, Cloudflare's cloud platform for running code, and does without Chromium and without a browser container. With this release, the company delivered on the promise it made in its August announcement.
What Kitesurf is and why agents need it
Kitesurf is a stateless browser: it does not keep state between sessions. It parses HTML, applies CSS, executes the page's JavaScript, and renders screenshots. From the outside, it looks like a regular browser. It supports the Chrome DevTools Protocol (CDP), the standard protocol that Puppeteer and most agent tools use to talk to browsers. That means you can connect to it the same way you would connect to Chrome.
According to the authors, the goal is to give agents a lightweight browser that uses less CPU and memory than Chromium and scales on demand across Cloudflare's global Workers network. The announcement contains no exact benchmarks. Cloudflare promises to publish them later, along with compatibility data against the Web Platform Tests.
The source code is on GitHub under the Apache 2.0 license. You can read it, modify it, run it locally, and deploy it to your own Cloudflare account. Until now, Kitesurf was available only as the managed Browser Run service.
How it works under the hood
The engine is built from open-source Rust libraries compiled to WebAssembly:
- Blitz handles HTML parsing, the DOM, layout, and rendering;
- Stylo processes CSS, cascades, and computed styles;
- Boa runs JavaScript inside WebAssembly, but only as a last resort, when the code cannot be executed in Dynamic Workers;
- the Linebender libraries handle text layout and 2D graphics, and Krilla generates PDFs.
The architecture has three parts: Engine manages browser sessions and client connections, PageScript runs the page and its JavaScript, and PageRenderer draws the result. All of this lives in separate folders under the packages directory, alongside the Rust core, the DOM layers, and the Kitesurf command-line tool. Third-party projects keep their own licenses, which are listed in the NOTICE file.
Each page gets an isolated Worker through Dynamic Workers. This is the foundation of the isolation, and it is also why self-deployment requires a paid Workers Paid plan.
How to try it
Running it locally is simple. Clone the repository, run pnpm install --frozen-lockfile, then pnpm run dev. The first build takes noticeably longer than the rest, because the Rust core and all its dependencies get compiled; after that, the cache speeds things up. Wrangler starts a local server at 127.0.0.1:8787. You can then open the address in Chrome DevTools, or connect from Node.js using puppeteer-core at ws://127.0.0.1:8787/devtools/browser, open a page, and take a screenshot. You can also point your agent straight at the local CDP port.
When your changes are ready, running pnpm run deploy from the repository root deploys the engine to your account, and you use it in place of the managed Browser Run APIs.
Limitations and contributing
The authors are upfront that the project is under active development. Current work focuses on compatibility with web standards, reliability, and efficiency. This means complex sites with non-standard JavaScript or heavy layouts may not render the way they do in Chromium. Before moving critical workflows to Kitesurf, test them against your target pages.
Third-party contributions follow a specific order: first open an issue with a minimal reproduction or a description of the approach, get a vouch from a maintainer, and only then submit a pull request. Small, targeted fixes with tests are easier to review than large rewrites. You can also help without writing code, for example by providing a page that reproduces a rendering bug or describing an automation scenario that fails.
What this means in practice
For teams building agents with web access, open-source Kitesurf is a reason to evaluate an alternative to a self-managed fleet of Chromium instances. It is especially relevant if you already run on Cloudflare and want the browser, isolation, and scaling in one platform. And the open license removes the dependency on the managed service: you can inspect the code, fork it, and adapt it to your own needs.
- Nov 21, 2026Figma AI Handled 18 of 19 Design System Tasks, From Tokens to Tables
- Nov 21, 2026Google Puts Hydrogen, a Model 2.5x Faster Than Gemini 3.8 Flash, Into Search
- Nov 21, 2026AcpGram Lets You Control Claude Code, Codex, and Gemini CLI Directly from Telegram
- Nov 21, 2026Bolt photo viewer for Mac advances to the next frame in 31 ms, ApolloOne takes 87
Key Questions Answered
What Kitesurf is and why agents need it?
Kitesurf is a stateless browser: it does not keep state between sessions. It parses HTML, applies CSS, executes the page's JavaScript, and renders screenshots. From the outside, it looks like a regular browser. It supports the Chrome DevTools Protocol (CDP), the standard protocol that Puppeteer and most agent tools use to talk to browsers. That means you can connect to it the same way you would connect to Chrome.
How it works under the hood?
The engine is built from open-source Rust libraries compiled to WebAssembly: Blitz handles HTML parsing, the DOM, layout, and rendering; Stylo processes CSS, cascades, and computed styles; Boa runs JavaScript inside WebAssembly, but only as a last resort, when the code cannot be executed in Dynamic Workers; the Linebender libraries handle text layout and 2D graphics, and Krilla generates PDFs.
How to try it?
Running it locally is simple. Clone the repository, run pnpm install --frozen-lockfile, then pnpm run dev. The first build takes noticeably longer than the rest, because the Rust core and all its dependencies get compiled; after that, the cache speeds things up. Wrangler starts a local server at 127.0.0.1:8787. You can then open the address in Chrome DevTools, or connect from Node.js using puppeteer-core at ws://127.0.0.1:8787/devtools/browser, open a page, and take a screenshot.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.