AI teaching assistants · officehour.ai

The AI TA your course actually controls.

Upload your materials, decide how the assistant answers, publish it to your class — and read every conversation it has. Two things it will never do: complete graded work, or answer from outside what you assigned.

Isolated materials & records
Every reply traceable to a config
Your files stay your files
What a student gets
Case Coach

I won’t write the analysis for you — that’s the part you’re assessed on. But four cost drivers matter here. Pick the one a funded entrant couldn’t copy in two years, and tell me why.

Lecture 4 · sl. 12Ryanair case · p. 6

Answers cite the document and page they came from. Students can export any conversation with a verification code, so an AI-assisted submission can be checked rather than argued about.

Three steps, about half an hour

No integration work. No IT ticket. Nothing to install for you or your students.

01
Upload what you teach

Slides, readings, cases — PDF, DOCX, PPTX. Indexing runs in the background and tells you when a file can't be read, instead of silently dropping it.

02
Set how it answers

Question-first or answer-first. Strict to your materials or flexible. Which documents each assistant may use. Test it privately before anyone sees it.

03
Share a join code

Students sign up themselves and agree to your review before their first question. Pause the assistants for exams whenever you need to. You read every conversation.

Built to refuse, not just to help

It won't complete graded work

A platform rule, above your settings and above anything a student types to talk it round. During assessment windows the Assessment Shield refuses the specific questions you protect.

It says when it doesn't know

On strict grounding it declines rather than improvising — and you see every refusal in the dashboard as a gap in your materials.

Exports can be verified

Each export carries a code you can check against the server — metadata only, no contents. No sign-in required, so a TA or second marker can use it. Verify an export

Every reply names its version

Change a setting mid-term and old conversations stay attributed to the configuration that produced them.

Students are told, in plain words, before they start

Monitoring isn’t buried in terms. Before their first question, students see exactly what is stored and who can read it — their instructor reads every conversation — and they have to agree. Research use is a separate, optional consent they can withdraw at any time — declining changes nothing about the course.

  • Train models on your materials or your students’ conversations, or let our AI provider keep them: they are sent with storage switched off
  • Share anything between courses, or with other instructors
  • Keep student records past the retention period stated in the policy: they are purged a set time after a course is closed, sooner if the instructor deletes it
  • Let students see each other's activity, ever

Deletion is real: an instructor who deletes a material or an archived course erases the files, the search index and the records, not just the pointer to them. A student who deletes a conversation removes it from their own view; the course record stays until the course is deleted or the retention period ends.

Read the full data-protection policy

Set it up this week, use it in next week’s class.

Request access with your institutional email. If your department has an invite code you’re in straight away; otherwise we confirm you’re teaching, usually within a working day.