Kitesurf: open source does not mean private by default.
Cloudflare opens its experimental agent browser. A deployment can expose the controls and the bill before authentication is added.

Concept photo: programming code on a screen, by luis gomes / Pexels. Illustrates programmable tools, not a Kitesurf product screenshot. Used under the Pexels license, which permits article, RSS and social reuse. No endorsement implied.
The browser starts, the deployment finishes, and the controls are already public. That is the uncomfortable interval in Kitesurf's own deployment instructions: an open-source engine can become an open browser endpoint before its operator adds authentication.
Cloudflare's November 13 announcement makes its experimental browser engine available under Apache 2.0. The developer changelog is dated November 12; November 14 coverage brings it to a wider audience. Those are different dates, not three launches.
The missing lock is documented
The current repository README warns that the engine has no authentication of its own. Someone with its URL can open DevTools, connect a Chrome DevTools Protocol client, send outbound requests through the operator's account, and use CPU time and Dynamic Workers that account pays for. The default public deployment address is predictable; an obscure name is not an access policy.
The ordinary instructions deploy first and protect the URL immediately afterward. The README explicitly identifies the gap between those steps. Its prevention route is to enable Access for all Workers in the account before deploying. Another authentication layer is also allowed by the changelog. This is a documented deployment boundary, not evidence that a particular installation has been exposed.
Preview is not a synonym for private, either. The repository says version URLs can be public too and need protection. A staging link deserves the same access check as the main endpoint.
A familiar connection is not full Chrome
Kitesurf parses HTML, applies CSS, runs page JavaScript and renders screenshots without Chrome or a browser container. CDP support lets clients such as Puppeteer connect. But the README separately says it is not a full Chrome implementation and some DevTools features do not work yet. Cloudflare calls the engine experimental and under heavy development.
That distinction matters before moving an existing workflow. A successful connection proves the client reached the engine. It does not prove the target site's forms, rendering or final result behave as they did in Chromium. Check the actual pages and outcomes your workflow needs, including its failure paths, before treating a migration as complete.
Open code, separate operating costs
Local development uses Wrangler. Running an instance on Cloudflare requires a Workers Paid plan because page isolation uses Dynamic Workers. Apache-licensed source does not make the hosted execution free. Dependencies retain their own licenses, too. AI Weekly's release summary also notes that the launch post supplies no CPU or memory figures for its efficiency framing; we are not turning that goal into a measured saving.
For a first evaluation, the sensible order is local compatibility work, a protected deployment route, and an access test before real browsing. If using Access with a programmatic CDP client, the README describes a service-token policy and connection headers. Keep that token secret: the documentation says it grants the same access as a signed-in user.
This is a documentation-led review, not a GigSoul deployment or performance test. The useful change is inspectable, modifiable browser code. The responsibility that stays with the operator is deciding who can drive it, before anyone else finds the wheel.


