Skip to content

Cloudflare open-sources Kitesurf, its browser engine for AI agents

Cloudflare's experimental browser engine for AI agents can now be inspected, run locally, or self-hosted on Cloudflare Workers.

Draftship Editorial3 min read

Cloudflare open-sources Kitesurf

The Kitesurf open source release is now official as of November 13, 2026. Cloudflare has made its experimental browser engine for AI agents available to inspect, run locally, or deploy as a self-hosted instance on Cloudflare Workers. The company introduced Kitesurf in August as a stateless browser aimed at AI agents and running wholly on Workers, then followed up in September with WebMCP.

Cloudflare's announcement is brief, but it is clear about access. The engine has moved from a Cloudflare-run experiment to code that builders can examine and run themselves.

What Cloudflare has confirmed

The confirmed details are narrow. Kitesurf is experimental. It is a browser engine for AI agents. It is now open source. Builders can read the code, run it on a local machine, or deploy their own instance on Cloudflare Workers. Cloudflare also describes Kitesurf as stateless and says it runs entirely on Workers. The post carries tags for Rust, WebAssembly and Workers. The announcement text provided does not include a license, a repository URL or setup instructions.

How the deployment model works

Only one part of the stack is described in the source: Workers. Kitesurf is meant to run there, and the open-source release lets a team deploy its own copy. For a builder, that changes the level of access. Instead of treating the browser engine as a black box, a developer can inspect the implementation before putting it near an agent workflow.

The source does not explain how local runs differ from Workers deployments, what configuration is required, or which interfaces an agent uses to drive the browser. It also does not say whether a Cloudflare-hosted version remains available alongside the open-source option.

What this likely changes for agent builders

Open source lowers the inspection barrier. A developer can check how the engine handles browser work an agent needs, run it locally for tests, and decide whether to deploy an instance on Workers. That matters when an agent has to interact with the web as part of a product.

But this is infrastructure, not an app feature. The release does not explain how to connect Kitesurf to a particular agent framework, model provider or application stack, and it does not promise compatibility with any of them. A team considering it would still need to do its own integration and testing work.

What remains unclear, and what this likely means

Key operational details are missing: license terms, where the code is hosted, documentation, release process, security model and support. Those gaps likely matter more for a browser engine than for a small library, because teams need to know how it is maintained and how it is isolated.

The September WebMCP follow-up is mentioned but not described in the excerpt. This suggests Kitesurf is part of a broader agent tooling effort, but the source does not connect the two in detail. Until the missing details are published, the sensible read is that Kitesurf is an experimental component for evaluation, not a settled default.

What this means for AI app builders

For teams adding AI to an app, Kitesurf is another sign that the agent layer is being split into components: models, tools and now the browser. Whether to adopt it depends on details the announcement leaves out.

Draftship, an AI app builder, has a separate guide on adding AI to an app for builders mapping out model calls and app logic. Cloudflare has not said that Kitesurf works with Draftship, and nothing in this release changes Draftship's features.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.