Cloudflare Launches Kitesurf, a New Browser Built for AI Agents

Cloudflare Launches Kitesurf, a New Browser Built for AI Agents

Cloudflare has launched Kitesurf, a new browser built specifically for AI agents that runs entirely on Cloudflare Workers. Available free in beta through Browser Run, Kitesurf is designed to consume substantially less CPU and memory than Chromium during common agent tasks such as taking screenshots and extracting HTML.

Cloudflare developed Kitesurf around the idea that browsers built for people carry features and resource requirements that AI agents often do not need. Conventional browsers such as Chromium have to support tabs, extensions, themes and highly polished visual rendering. Agents have different priorities, including efficient access to web content, scalability, machine-readable output and isolation from security threats. Cloudflare also points to emerging risks such as prompt injection and unsafe tool use as reasons to rethink the browser architecture around AI.

The project took shape over 12 weeks, with its first commit made in May. Cloudflare said improvements across its Workers platform — including WebAssembly support, Dynamic Workers, Durable Objects, RPC and increased Node.js compatibility — made a more ambitious browser possible. Growth in Browser Run also contributed to the decision, as AI agents increasingly depend on browser automation to interact with websites and complete tasks.

Rather than placing Chromium underneath the service, Cloudflare built Kitesurf around three primary components: Engine, PageScript and PageRenderer. The Engine serves as the browser’s public-facing layer, managing Chrome DevTools Protocol connections, REST APIs and session state. Support for CDP also means existing tools such as Puppeteer, Playwright and chrome-remote-interface can connect to Kitesurf without requiring an entirely new control system.

PageScript handles individual pages and out-of-process iframes using Dynamic Workers. It maintains the page environment and DOM, while Rust-based components process HTML and CSS. JavaScript and WebAssembly associated with the page execute inside the same isolate. For JavaScript eval calls, which Workers does not currently support natively, Cloudflare uses the Rust-based Boa JS engine as an additional runtime.

PageRenderer handles the visual side of the process, turning computed page information into outputs including JPEG, PNG and PDF. Cloudflare designed the renderer to remain stateless, allowing it to be discarded and restarted when a request stalls or fails rather than preserving unnecessary state between jobs.

Network access is isolated as well. A component called SandboxOutbound is the only part of Kitesurf permitted to directly retrieve external resources, and Cloudflare uses it to enforce policies including CORS handling, response filtering and separate cookie storage for each page. The broader architecture assumes every page an agent opens could contain untrusted input, so individual components receive only the access required for their specific jobs.

Kitesurf already passes more than 215,000 Web Platform Tests, according to Cloudflare, with hundreds of additional tests being added each week. The company says areas particularly relevant to agents — including HTML, CSS, DOM, SVG, selection and XHR — already have substantial coverage.

The biggest difference from Chromium appears in resource consumption. Across Cloudflare’s 14-URL benchmark, Kitesurf used 380 milliseconds of CPU time to produce screenshots, compared with 1,173 milliseconds for Chromium. HTML extraction consumed 229 milliseconds of CPU time with Kitesurf versus 877 milliseconds with Chromium.

Memory usage showed an even wider gap. Kitesurf consumed 57.8 MiB when generating screenshots, compared with 271 MiB for Chromium, while HTML extraction required 39.4 MiB versus 273.7 MiB. Those results translate to between 3.1 and 3.8 times lower CPU consumption and between 4.7 and 7 times lower memory use in the tasks Cloudflare tested.

That efficiency comes with a performance trade-off. Chromium completed screenshot tasks in 637 milliseconds of wall time, while Kitesurf took 1,148 milliseconds. HTML extraction took 472 milliseconds with Chromium and 820 milliseconds with Kitesurf. Cloudflare said much of the difference comes from rasterization and image encoding, areas it plans to continue optimizing.

The company is positioning Kitesurf primarily for workloads where agents need to interact with or extract information from web pages without requiring every capability of a conventional browser. That includes content extraction, screenshots, PDF generation and short-lived automation jobs where lower resource requirements can allow more simultaneous sessions.

Kitesurf is not yet suitable for every browser workload. It currently lacks support for video playback and WebGL, and Cloudflare says developers should continue using Browser Run’s Chromium option for bot challenges that require real TLS fingerprints or lengthy authenticated sessions that depend on persistent state.

Developers can access Kitesurf through Browser Run’s existing CDP and Quick Actions endpoints by selecting it as the browser option. Cloudflare has also released a public playground where developers can load websites and use Chrome DevTools to inspect how Kitesurf handles the DOM, network activity, console output and memory consumption. The browser can currently render sites and applications including Wikipedia, Hacker News, the Cloudflare Blog and multiple TodoMVC implementations.

Cloudflare plans to continue expanding CDP compatibility, Web Platform Test coverage, rendering accuracy and efficiency as the beta develops. The company also intends to open source Kitesurf once it is ready, with the eventual goal of allowing customers to deploy their own versions within their Cloudflare accounts.

Kitesurf is therefore less an attempt to recreate everything Chromium does than to narrow the browser around what AI agents actually require. Cloudflare is accepting some limitations in compatibility and wall-clock speed in exchange for substantially lower CPU and memory consumption, stronger isolation and an architecture designed to scale short-lived browser sessions for AI workloads.

This analysis is based on reporting from Cloudflare.

Image courtesy of Cloudflare.

This article was generated with AI assistance and reviewed for accuracy and quality.

Last updated: August 7, 2026

About this article: This article was generated with AI assistance and reviewed by our editorial team to ensure it follows our editorial standards for accuracy and independence. We maintain strict fact-checking protocols and cite all sources.

Word count: 907Reading time: 0 minutes
Browse All Articles
Share this article:
Next Article

AI News Daily

Breaking Intelligence • Since 2023

Join hundreds of thousands of AI professionals who start their day with our curated newsletter. Get breaking news, expert analysis, and exclusive insights.

Stay Ahead of AI

Get the latest AI breakthroughs, tools, and insights delivered to your inbox every week.

Free forever Unsubscribe anytime No spam guarantee

Go Premium

Unlock unlimited AI tools and an ad-free reading experience designed for AI professionals.

• Ad-free experience• Premium AI tools
Start Free Trial

14-day free trial • Cancel anytime
Plus $9/mo • Pro $90/yr (2 months free)

Follow Our Community

ChatAI

Breaking Intelligence

Your daily briefing on what matters in AI. Trusted by developers, researchers, executives, and AI enthusiasts worldwide.

© 2026 ChatAI. All rights reserved.