Cloudflare has launched Kitesurf, a browser designed specifically for artificial intelligence agents rather than human users. The new browser runs entirely on Cloudflare Workers and is built to reduce the computing and memory overhead associated with conventional browsers such as Chromium, potentially making large-scale AI web automation more practical and economical.
Kitesurf is being introduced as part of Cloudflare’s broader push into infrastructure for autonomous AI agents. The browser is currently available for free in beta through Cloudflare’s Browser Run service, where developers can use existing browser automation tools and AI agents with Kitesurf. The company says it is targeting workloads where agents need to retrieve web content, take screenshots, interact with pages or perform automated tasks without requiring the full capabilities of a human-oriented browser.
What Happened
Cloudflare unveiled Kitesurf on August 6 as an “agent-first” browser built on its Workers platform. The project emerged from the company’s observation that conventional browser engines carry significant overhead that is unnecessary for many AI workloads.
Browsers such as Chromium are designed around human interaction, including features such as tabs, extensions, high-fidelity visual rendering and smooth scrolling. AI agents have different requirements. They need access to web pages, structured content, browser controls and scalable execution, but do not necessarily require the complete visual experience designed for people.
Cloudflare said Kitesurf is currently in beta and can be accessed through Browser Run. The company plans to continue improving compatibility and performance before moving toward production readiness.
Kitesurf at a Glance
| Category | Details |
|---|---|
| Product | Kitesurf |
| Developer | Cloudflare |
| Designed For | AI agents and web automation |
| Infrastructure | Cloudflare Workers |
| Runtime | V8 isolates |
| Current Status | Free beta |
| Browser Automation | Browser Run |
| Compatibility | CDP, Puppeteer, Playwright |
| Web Platform Tests | More than 215,000 |
| Open-Source Plan | Cloudflare intends to open source it |
Why Cloudflare Built a Browser for AI Agents
AI agents increasingly need browsers to perform tasks on the web. An agent may need to search websites, extract information, fill forms, interact with applications or capture screenshots before completing a larger task.
Running a conventional browser instance for every agent can become expensive because browsers consume significant memory and CPU resources. This becomes particularly important when thousands or millions of short-lived agent tasks need to run simultaneously.
Cloudflare’s solution is to remove features that are primarily useful to humans and optimize the browser around the needs of machines.
The company says AI agents care more about token consumption, context, scalability, performance and cost than features such as browser themes, extensions or perfect visual rendering.
Performance Compared With Chromium
Cloudflare’s testing shows that Kitesurf uses substantially less CPU and memory than Chromium for several common agent workloads.
The company compared five Browser Run quick-action runs across a 14-URL test corpus. While Chromium was faster in wall-clock time, Kitesurf required significantly fewer computing resources.
| Test | Kitesurf | Chromium | Kitesurf Advantage |
|---|---|---|---|
| CPU — Screenshot | 380 ms | 1,173 ms | 3.1x less CPU |
| CPU — HTML Extraction | 229 ms | 877 ms | 3.8x less CPU |
| Memory — Screenshot | 57.8 MiB | 271.0 MiB | 4.7x less memory |
| Memory — HTML Extraction | 39.4 MiB | 273.7 MiB | 7x less memory |
| Wall Time — Screenshot | 1,148 ms | 637 ms | Chromium faster |
| Wall Time — HTML Extraction | 820 ms | 472 ms | Chromium faster |
Cloudflare says the lower CPU and memory requirements are particularly important because those resources directly influence infrastructure costs and the number of browser sessions that can be operated simultaneously.
Built on Cloudflare Workers
Kitesurf runs entirely on Cloudflare Workers, using technologies including WebAssembly, Dynamic Workers, Durable Objects and Worker-to-Worker RPC.
The browser uses V8 isolates to separate components and workloads. Cloudflare has also designed Kitesurf around statelessness, allowing individual components to be discarded and restarted rather than maintaining expensive persistent browser infrastructure.
This architecture is particularly suited to AI workloads, which can be highly variable. An agent might launch hundreds of browser tasks simultaneously and then stop using them once a particular job is complete.
Instead of maintaining always-on browser instances, Kitesurf can create resources as required and dispose of them when they are no longer needed.
Security and Isolation
Security is another major consideration because AI agents can be directed toward arbitrary websites containing potentially hostile content.
Cloudflare designed Kitesurf with the assumption that every web page should be treated as untrusted input. Each session starts fresh, and components are isolated so that they only receive the resources required for their specific functions.
The browser also uses a dedicated outbound networking component to control how pages retrieve external resources. Cloudflare says this component handles policies including CORS, browser-style headers, response filtering and separate cookie storage.
This architecture is particularly relevant for agentic browsing because prompt injection and malicious web content can potentially influence an AI system while it is navigating the internet.
Compatibility With Existing Developer Tools
Cloudflare has attempted to make Kitesurf compatible with existing browser automation infrastructure rather than requiring developers to build new integrations from scratch.
The browser exposes the Chrome DevTools Protocol (CDP), allowing tools such as Puppeteer, Playwright and Chrome DevTools to interact with it.
Developers using Cloudflare’s Browser Run service can select Kitesurf by specifying the browser option in the relevant endpoints. AI agents that support MCP and CDP can also connect to the browser.
This compatibility could make adoption easier for developers already building browser-based AI agents.
Kitesurf Passes More Than 215,000 Tests
Cloudflare says Kitesurf currently passes more than 215,000 Web Platform Tests, with hundreds of additional tests being added each week.
The company has focused particularly on web technologies that are important for agent workloads, including HTML, DOM, CSS, SVG, selection and XHR. It also uses integration and visual regression testing against real websites to compare Kitesurf with Chromium.
However, Kitesurf does not yet offer complete compatibility with every website or browser workload.
Current Limitations
Cloudflare acknowledges that Kitesurf is not intended to replace Chromium in every situation.
The browser currently is not the preferred choice for workloads requiring:
- Video playback
- WebGL
- Certain bot-challenge handshakes
- Long authenticated sessions requiring persistent state
- Full pixel-perfect rendering
For these workloads, Cloudflare recommends continuing to use the Chromium-based default in Browser Run.
The distinction is important because Kitesurf is designed as a specialized tool rather than a general-purpose consumer browser.
Why It Matters for AI Agents
The launch reflects a broader change in computing infrastructure as AI systems become capable of taking actions rather than simply generating responses.
Cloudflare has already been expanding its Agent Cloud infrastructure, including tools for building, deploying and scaling autonomous agents. In April, the company said its platform was being designed to move agents from experiments on local machines toward production workloads operating across its global network.
Kitesurf extends that strategy into web browsing.
Instead of treating the browser as an application that a person opens, Cloudflare is treating it as infrastructure that an AI agent can temporarily invoke whenever a task requires web access.
Industry Impact
The development could encourage other infrastructure providers to rethink the traditional browser stack for AI workloads.
As agents become more common, developers may increasingly prioritize low-cost, high-concurrency browser sessions over the complete feature set required by human users. This could create demand for specialized browsers optimized around machine-readable content, automation, isolation and resource efficiency.
It also highlights a growing distinction between human-facing software and agent-facing infrastructure. The same web may increasingly need to support two types of users: people who require rich visual interfaces and AI systems that prioritize structured information and efficient execution.
Looking Ahead
Cloudflare is still treating Kitesurf as an early-stage project, but its development points toward a potentially important shift in how web infrastructure is built for AI agents. The company plans to improve CDP support, rendering quality, Web Platform Test coverage and overall efficiency before the browser is considered production-ready. Cloudflare has also said it intends to open source Kitesurf, potentially allowing customers to deploy their own versions on their Cloudflare accounts.
The bigger question is whether specialized agent browsers can become a standard layer in the emerging agentic internet. Developers will be watching the balance between compatibility and efficiency, while businesses will evaluate whether lower resource consumption translates into meaningful cost savings at scale. If AI agents increasingly become autonomous users of the web, products such as Kitesurf could become as important to agent infrastructure as conventional browsers have been to human internet access.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
