Run an essay exam you can stand behind.

Four things, during the session only: the names of running applications, checked against the sharing and remote-control list; the number of attached displays; virtual-machine signals; and, where the network check runs, whether the student's public address matches yours. The sweep runs once before entry and about every five seconds while the student writes. Weft also saves the essay itself as the student types it, which is how the work reaches you for grading.

The camera: the macOS app contains no webcam capture, so nothing is photographed unless a school explicitly enables such an option in a future deployment. Audio: never recorded. Keystrokes: Weft saves the essay text in the exam editor and nothing else; there is no keylogger, and nothing a student types outside the exam is ever seen. Files on the machine: never read. The screen as pictures: the current app takes no screenshots and records no video. And anything at all after the session closes: monitoring ends the moment the student submits or you end the session.

Essays and their autosaved drafts, check results, outlines, and grades are rows in a Postgres database hosted by Supabase. Reference files sit in a private storage bucket and are fetched through short-lived signed links. Every read and write is tied to the signed-in school Google account by row-level security in the database itself, not merely by the interface.

You see the sessions you ran: rosters, check results, essays, outlines, and the grades you write. A student sees their own work, and a grade only after you share it; the release email tells them a grade is ready and never contains the score.

Yes, twice. Before entering, the checks screen shows the same two columns this page does: what it checks, and what it never does. After submitting, a data ledger lists exactly what was captured during their session and notes that monitoring stopped at submit.

Almost nothing, automatically. A detected sharing app pauses writing until it closes; everything else lands as a status in your roster. Weft never ejects a student, never deducts points, never accuses. Only a concrete finding is flagged, an unperformed check is never a warning, and every flag travels with its benign explanation, because a false accusation harms a real student more than a missed borderline case ever would.

The editor autosaves as they type, and the save label tells the truth: if a save fails, it says so and retries until one lands. Submitting requires a successful final save; if the network is down at that moment, the student stays in the locked exam with their work intact while Weft keeps retrying. Work is never discarded to make an error message go away.