Kitesurf explained: Cloudflare's browser built for AI agents, not humans

Cloudflare's Kitesurf strips away browser features meant for human users to reduce compute and memory requirements for AI agents performing web tasks such as extraction, screenshots and PDF generation

 

Content delivery network service provider Cloudflare has launched Kitesurf, a cloud-hosted browser designed specifically for artificial intelligence (AI) agents rather than human users. According to the company, Kitesurf is designed to help AI agents navigate websites, extract information, take screenshots and complete web-based tasks while using less computing power and memory than conventional browsers such as Chromium.

 

Kitesurf runs on Cloudflare Workers and strips away features that human users typically require, such as tabs, extensions and high-fidelity rendering. Instead, it focuses on requirements specific to AI agents, including context windows, token usage, performance, memory consumption and cost.

 

The browser is available through Cloudflare's Browser Run platform, which allows AI agents to interact with websites and perform actions through browser automation. Browser Run itself supports tasks such as navigating webpages, extracting rendered content, generating PDFs and capturing screenshots.

 

What is Cloudflare Kitesurf

Kitesurf is designed as a lightweight, cloud-based browser for AI agents and automated workflows.

 

Unlike conventional browsers, which are built to render rich interfaces for people, Kitesurf is optimised for machine-driven tasks in which an AI system needs to load a webpage, read its content or perform an action.

 

It runs on Cloudflare's serverless Workers infrastructure and is designed to create isolated browser sessions for individual AI tasks rather than maintaining long-running sessions.

 

This makes it particularly suited to short, bursty workloads where a browser session can be created for a task and discarded once the work is complete.

 

The approach also allows Cloudflare to remove browser features that AI agents may not require and focus instead on reducing compute and memory requirements. 

How Kitesurf compares with Chromium

Cloudflare's benchmarks suggest Kitesurf can use considerably fewer computing resources than Chromium for some common agentic workloads.

 

According to the company's tests, Kitesurf used 3.1 times less central processing unit (CPU) resources while taking screenshots and 3.8 times less while extracting HTML.

 

Memory consumption was also lower. Kitesurf used 4.7 times less memory for screenshots and seven times less for HTML extraction.

 

Chromium, however, completed the tests faster in terms of overall wall-clock time.

 

The comparison illustrates the trade-off behind Kitesurf. It is not designed to replace a fully featured browser in every situation. Instead, Cloudflare is targeting workloads where lower infrastructure requirements and the ability to scale large numbers of temporary browser sessions matter more than maximum rendering performance.

Built for AI agents

Kitesurf can work with Browser Run's Quick Actions, Chrome DevTools Protocol (CDP) and Model Context Protocol (MCP) clients.

 

Browser Run already allows developers to use Quick Actions for stateless tasks such as capturing screenshots, generating PDFs, fetching rendered HTML and extracting structured information.

 

For more complex automation, developers can use CDP to control browser sessions directly. Cloudflare also supports MCP clients through CDP, allowing compatible AI coding agents to navigate webpages, take screenshots and perform other browser automation tasks.

 

This means developers can integrate Kitesurf into existing AI-agent workflows rather than building browser infrastructure from scratch.

 

Cloudflare said Kitesurf was developed in about 12 weeks.

 

Its technology combines several existing components, including the Blitz rendering engine, Firefox's Stylo Cascading Style Sheets parser and the Boa JavaScript engine, with the rest of the system running on Cloudflare Workers.

What Kitesurf can be used for

Kitesurf is intended for AI agents and automated workloads where a full Chromium browser may not be necessary.

 

It can be used to render webpages, extract content, generate PDFs and capture screenshots through Cloudflare's browser tools.

 

Cloudflare's Browser Run Quick Actions already support tasks such as converting webpages into Markdown, extracting structured JavaScript Object Notation data, scraping specific page elements and retrieving links.

 

Kitesurf is particularly suited to short-lived tasks because each browser instance can operate independently and be discarded after the task is completed.

 

Cloudflare said the browser can render websites including Wikipedia, Hacker News and the Cloudflare Blog, as well as web applications built using frameworks such as React, Vue and Angular. 

Where Kitesurf falls short

Kitesurf is not intended to replace Chromium for every browser workload. It currently lacks support for features including video playback and WebGL. It also cannot handle some bot-challenge processes that depend on conventional browser Transport Layer Security fingerprints.

 

The browser is also not designed for long-running workflows where users need to remain authenticated and maintain persistent browser state. For such workloads, Cloudflare recommends using the standard Browser Run option powered by Chromium.

 

Browser Run supports full browser sessions through tools including Puppeteer, Playwright and CDP, alongside simpler stateless Quick Actions.

Kitesurf is currently in beta

Kitesurf is currently available in beta through Browser Run, subject to account limits.

 

Cloudflare's Browser Run platform itself is available on free and paid plans. The free tier currently includes limited daily browser usage, while paid plans provide higher usage and concurrency limits.

 

The launch reflects a broader shift in browser design as AI agents become increasingly capable of interacting directly with the web.

 

Traditional browsers were designed around human behaviour: opening tabs, rendering rich interfaces and maintaining long browsing sessions. AI agents have different requirements.

 

Kitesurf represents Cloudflare's attempt to redesign the browser around those machine-driven workloads, prioritising lower compute costs, faster scaling and simpler automation over features meant primarily for human users.