This is a static application. There is no server component, no account, and no synchronisation service, and that shapes both what it can promise and what it cannot do for you.
What runs offline
After the first load, the workbench, its code, styles, icon, sample data and method pages are served from a local cache. Profiling, comparison, disposition, freezing and every export run entirely in the browser. The only network request the product makes at any point is the service worker fetching its own files from its own origin; there is no analytics, no telemetry, and no application programming interface call.
What never leaves the device
File contents are read in the page and never uploaded. What the project stores is structural: column names, positions, inferred types with confidence, blank counts and rates, bounded distinct counts, length ranges, a content digest, and the notes and policies you wrote. Source rows and raw example values are not retained, so they cannot appear in a checkpoint or a packet either.
What that costs you
There is no cloud backup and no recovery. Clearing site data, using a private window, switching browsers, or a browser evicting storage under pressure all remove the local project. Export a checkpoint at the end of every session and keep it wherever your other work records live.
There is also no collaboration layer. Two reviewers cannot edit the same project at once; a second tab is read only and says so, with an explicit control to take over the writer lease. Sharing happens by sending a packet or a checkpoint, deliberately, as a file.
What you remain responsible for
Device security, disk encryption, who can open the browser profile, where checkpoints are stored, and whether the files you profile are yours to process. An account-free product cannot enforce any of that for you; it can only avoid becoming another place where the data lives.