Cloudflare unveils Kitesurf, a browser built for AI agents

Cloudflare has launched Kitesurf, a cloud-hosted browser designed for AI agents rather than human users. The move matters because it reflects a broader shift in artificial intelligence: software is moving beyond chat to systems that can browse the web, fill out forms, and carry out tasks on a user’s behalf.

Unlike consumer browsers built around tabs, themes and extensions, Kitesurf is optimized for speed, cost, scalability and security in machine-to-machine workflows. Cloudflare says the product is already available in beta through Browser Run, where developers can programmatically control headless browser instances across its network.

Why Cloudflare is betting on an AI-agent browser

Cloudflare’s launch is part of a fast-growing race to build the web infrastructure AI agents will need. As large language models become more capable, the next step for many developers is not just answering questions but completing actions: logging in, searching, comparing, extracting data and submitting requests on the web.

That evolution creates a technical problem. Traditional browsers were not built for software agents. They are optimized for people, which means they spend resources rendering visual interfaces, managing tabs, loading extensions and handling user-facing features that automated systems often do not need.

Cloudflare argues that an AI-first browser should be designed around the realities of agentic workloads. Those include context-window limits, token efficiency, throughput, memory use and the ability to run at scale across many concurrent tasks.

There is also a security challenge. AI browsers can be more exposed to attacks such as prompt injection, where malicious web content tries to manipulate a model’s behavior. That threat model is different from the one browser makers have traditionally focused on, and it helps explain why companies are exploring purpose-built tools instead of repurposing existing browsers.

How Kitesurf works

Kitesurf is a cloud-hosted browser that developers can use to build AI agents capable of interacting with websites in a browser-like environment. Cloudflare says the software can navigate pages, complete forms and support other browser-based actions without requiring teams to build and maintain their own browser stack.

The product is currently free during beta, and Cloudflare is offering access through Browser Run, a service that lets developers control headless browser instances on the company’s network. In practical terms, that means the browser is running in Cloudflare’s infrastructure rather than on a user’s local machine.

Cloudflare says the approach is meant to cut compute requirements compared with using Chromium-based workflows. The company’s pitch is that developers can run agent tasks more efficiently and keep costs lower, particularly for repetitive operations like extracting HTML or taking screenshots.

What Cloudflare says is different about Kitesurf

Cloudflare says Kitesurf is tuned for machine use, not visual polish. That means elements many people associate with a browser experience—such as themes, tabs and extensions—are not the priorities here. Instead, the emphasis is on performance and reliability for automation.

The company says Kitesurf is more efficient in CPU and memory use than Chromium for common agent tasks. For businesses trying to deploy AI at scale, those savings can matter as much as raw speed. Lower resource usage can translate into lower operating costs and fewer infrastructure bottlenecks.

Cloudflare says the browser uses less CPU and memory than Chromium for common agentic jobs such as screenshots and HTML extraction, making it a more efficient option for developers building automation tools.

What is under the hood?

Kitesurf is not a from-scratch browser in the traditional sense. Cloudflare says it assembled the system from several existing technologies and wrapped them inside its serverless platform, Workers.

The company says the browser incorporates a modular rendering engine from Blitz, Mozilla’s Stylo CSS parser and Boa, a Rust-based JavaScript engine. Everything else runs within Cloudflare Workers, which allows the browser to take advantage of the company’s distributed infrastructure.

That design choice matters because it positions Kitesurf less as a standalone consumer app and more as a cloud-native automation layer. In effect, Cloudflare is treating the browser as part of its developer platform rather than as a product meant to compete head-on with Chrome or Safari.

How quickly did Cloudflare build it?

Cloudflare says the browser was conceived only 12 weeks ago, an unusually fast timeline for a web platform product. That speed suggests the company sees agentic browsing as an urgent opportunity rather than a long-term research project.

Speed also reflects the state of the broader market. As more AI companies ship agents and browser-based automation, infrastructure providers are trying to establish early technical standards and developer loyalty. Cloudflare’s decision to move quickly may help it shape the tools developers adopt before the market consolidates around a single approach.

The beta also shows Cloudflare is willing to iterate in public. Rather than waiting for a polished final release, it is letting developers experiment now, while the company continues to add tests and improve compatibility.

What can Kitesurf already do?

Cloudflare says Kitesurf already passes more than 215,000 web platform tests, and that the number is growing by hundreds every week. The company is using those results to signal that the browser is maturing quickly even though it is still early in development.

Cloudflare says the browser correctly renders a range of familiar sites and benchmarks, including TodoMVC, Wikipedia, Hacker News, the Cloudflare Blog and much of the company’s own dashboard. That matters because browser automation tools are only useful if they can behave consistently across a wide variety of real-world pages.

The company also emphasized that the browser can support common benchmark-style tasks, such as retrieving screenshots and extracting HTML. Those are basic building blocks for many AI workflows, from document processing to lead generation and research automation.

Why benchmark support matters

Benchmark support matters because developers need proof that an automation tool works beyond a controlled demo. If an AI browser can render standard applications and handle common sites reliably, it becomes easier to build production software on top of it.

In AI infrastructure, reliability often matters more than novelty. A browser that succeeds 90% of the time in a toy example is not enough if enterprise workflows depend on accuracy across thousands of requests. Cloudflare appears to be using test coverage as a credibility signal for potential customers.

Who inspired Kitesurf?

Cloudflare says the open-source Rust project Obscura helped inspire the browser’s development. The company noted that the first proof of concept was a port of Obscura to Workers, suggesting the product grew out of experimentation rather than a top-down product plan.

That origin story is notable because it places Kitesurf in the broader open-source ecosystem that underpins much of modern web infrastructure. Cloudflare is not only repackaging existing browser technology; it is also acknowledging the role of community-led engineering in pushing the category forward.

Cloudflare said its initial proof of concept came from adapting the open-source Obscura headless engine to run on Workers, which helped shape the direction of Kitesurf.

Why browsers matter in the AI agent era

Browsers are becoming central to the next phase of AI because so much of the digital economy still happens on the web. Even as APIs, app integrations and enterprise connectors expand, many essential tasks remain browser-native: booking travel, processing forms, checking dashboards, scraping public pages and interacting with tools that lack structured interfaces.

That means the browser may become one of the most important layers in agent infrastructure. If a model can reason but cannot reliably interact with websites, its real-world usefulness is limited. A browser built for agents helps close that gap.

The shift also raises strategic questions for major cloud and infrastructure companies. If agent traffic grows, whoever owns the browser layer could shape how AI systems discover, access and use web content. That could affect both technical standards and commercial relationships across the internet.

How Kitesurf compares with Chromium-based tools

Cloudflare’s comparison with Chromium is telling. Chromium has become the default base for many automation tools because it is familiar, mature and broadly compatible. But that does not make it ideal for every workload.

Kitesurf is aimed at reducing overhead in cases where the full browser experience is unnecessary. For developers running large numbers of agent tasks, shaving CPU and memory use can improve throughput and reduce infrastructure costs. In a cloud setting, those gains can be significant.

At the same time, Chromium remains deeply entrenched, and it will likely continue to power many browser automation systems. Kitesurf’s success will depend on whether its efficiency advantages outweigh the engineering effort required to adopt a new platform.

• Cloudflare launched Kitesurf as a cloud-hosted browser for AI agents.
• The browser is currently in beta and available through Browser Run.
• Cloudflare says it uses less CPU and memory than Chromium for common agent tasks.
• Kitesurf is built on Workers and combines multiple open technologies.
• The product is designed for browsing, form filling and other web automation.

What does this mean for developers?

For developers, Kitesurf could remove a major piece of engineering work from the AI stack. Instead of assembling and maintaining a browser environment themselves, teams can use Cloudflare’s platform to run agent-facing browser tasks at scale.

That could be especially appealing for startups and internal tools that need to automate websites quickly without investing in browser infrastructure expertise. It may also help teams prototype agents faster, since the browser layer is already hosted and integrated with Cloudflare’s network.

Still, the value proposition will depend on adoption. Developers will want to know whether Kitesurf can handle complex sites, authentication flows, dynamic content and the unpredictable behavior of live web pages. Beta access gives Cloudflare a chance to prove that reliability before a broader rollout.

What are the risks and open questions?

The biggest open question is whether a browser built for agents can remain secure while also being flexible enough for real-world use. Prompt injection is one concern, but there are others too, including abuse, access control, session handling and the risk of agents taking unintended actions when web content is malformed or deceptive.

Another question is whether a cloud-native browser can become a durable standard or simply another tool in a crowded automation market. Cloudflare is entering a space that may eventually attract browser makers, AI labs, cloud vendors and open-source projects all at once.

There is also the broader issue of web economics. If AI agents increasingly browse the web at scale, publishers, platforms and infrastructure providers may need to rethink traffic patterns, bot detection, authentication and pricing. Kitesurf is not just a product launch; it is a sign of how the web itself may evolve.

Timeline: how Kitesurf reached beta

The product moved from idea to beta in a compressed period, which is unusual even by AI industry standards. Here is the sequence Cloudflare described:

The bigger picture: browsers are becoming AI infrastructure

Cloudflare’s launch underscores a broader industry trend: the browser is no longer just an end-user app. In the age of AI agents, it is turning into a programmable infrastructure layer that can mediate between models and the public web.

That change could create new competition among cloud providers, open-source projects and browser vendors. It could also shift attention away from consumer-facing features and toward engineering priorities like observability, sandboxing, throughput and security.

For now, Kitesurf is still in beta and aimed squarely at developers. But the launch signals that Cloudflare believes the next major browser category will not be built for people at all. It will be built for software that behaves like a user, only faster, cheaper and at much larger scale.

If that prediction proves right, the browser race may be entering a new phase—one where the most important customers are not the people clicking links, but the AI systems doing the clicking for them.