It stays on your computer
Feedbacker has no accounts and no server. It runs on your own computer, and your work lives in folders you keep: on your own disk, or on storage your institution allows. Feedbacker doesn't host your work: there is no Feedbacker server holding it, and I never see it. What you approve and send is processed by the AI provider (see below).
Names are replaced before anything is sent
Students' names, IDs, emails, links and phone numbers are replaced with tokens on your computer, and the key that links the tokens back to the students never leaves it. You read each text, see what every token replaced, and approve it before it can be sent. Change anything afterwards and the approval is cleared.
Strictly, this is pseudonymisation, not anonymisation: because you hold the key, the work is still personal data, and Feedbacker treats it that way.
The replacing follows rules, so it can miss a name nobody listed, or a clue such as an employer. That is why you review every text. Images can't be redacted, so you look at each one and choose whether to send it. The hidden details inside image files, such as the camera, the date, the place and the author, are removed.
You decide what the AI sees
Nothing goes to the AI until you have approved the material and confirmed the request, having seen the most it could cost. Only what the task needs is sent. For a reading: the approved text, the rubric, the brief if you include it, and any images you approved. For feedback drafts: the approved text of that one student's submission, the rubric, the brief if you include it, your approved feedback guide if you have one, and your own levels, marks and comments for that student. Never another student's work, the original files, the names or the key.
Requests go to Anthropic's Claude with your own API key, through a small program on your computer that checks each one before it leaves. What you send is processed by Anthropic, under its own terms, and some of it is kept for a time: part of a request (the instructions, the rubric, the brief and the feedback guide, never a submission) can be held for five minutes to lower the cost, and the results of a batch are kept for 29 days so you can collect them. Its terms on training and retention are for you or your institution to check: For institutions lists what to confirm.
A record of what was sent
A log on your computer records every request: when, which AI, a fingerprint of what was sent, and the cost. It holds no text and no names, so you can show it to someone who is checking without showing them students' work. Every suggestion records which AI made it and what it was sent, and feedback you adapt from an AI draft is labelled as such.
You choose how long it is kept
Each piece of work records how long its material may be kept (90 days by default), for example under your institution's terms. Deleting a workspace removes the whole folder, and everything in it, in one action, once you have seen what it will take and typed its name. Feedbacker doesn't yet remind you when the time has passed: deleting is your action.
Open to inspection
The code, and the instructions the AI is given, are open source, so anyone can read what Feedbacker does. Data handling in full is the complete reference.
What this page is, and isn't
It describes what Feedbacker does. It is not a claim of legal, regulatory or institutional compliance: your institution makes its own assessment, and For institutions is written to support it. Before you use real students' work, confirm with whoever governs the assessment that AI assistance is acceptable.