A little application for a surprisingly slippery question.

Updated October 15, 2026 · Independent experiment

What does a browser agent bring back after you close it? CarryLab makes that question visible through a tiny app with seven different kinds of memory.

Why a desk?

A storage checklist can tell you whether an API exists. It is less useful for understanding what a user loses. A lamp, draft, notebook bag, guest pass, ink bottle, saved postcard and cached map give each storage mechanism a separate consequence. Click one object, make a controlled change to your browser session, then come back and observe it.

The scene is intentionally small. It uses generated fixture strings instead of real drafts, account cookies or uploaded state files. You can reproduce a browser-tool behavior without handing a test site your own authentication data.

Who it is for

Developers and testers working with agent-browser can compare a running session, serialized state and a dedicated profile. The ordinary reload and independent-tab recipes also work as a small browser-storage lesson. The native CLI results name the exact environment; the site does not promise that every automation tool saves the same data.

CarryLab is free to use. It runs on Cloudflare with browser-local fixtures and one same-origin endpoint for a fabricated HttpOnly cookie. There is no model execution, paid generation service, subscription or checkout.

Independence and related products

CarryLab is an independent project, not an official Vercel, agent-browser or browser-vendor service. It does not represent a previous organization or operator associated with this domain. Product names identify the tools being tested.

The own-application field note links to related WebTraceCheck and CleanTempMail services when a fresh browser or disposable staging inbox fits the next step. Those links are promotional and carry referral tags. They are optional and do not alter the experiment's results.

Corrections

If a repeatable result differs, keep your CLI and browser versions, exact operation, trial observation and sanitized command response. Send a concise reproduction to support@cleantempmail.com. Do not send real browser state files or credentials. Our method explains which conclusions the evidence supports, and the privacy note lists the exact fixture names.