Skip to content
Platform

An AI tutor that knows when to stop tutoring.

If a student says something that may mean they are in danger, Zelvi stops, says it is an AI, gives the right helpline and tells the people your school chose.

Talk to usSee it with your material
Student safety
A student types a worrying message. Zelvi stops tutoring, gives a fixed reply with crisis resources, alerts the school's safeguarding contacts, and shows the teacher a flag to check in.
Zelvi reads what students type and say for signs of danger, including slang and coded words.

  • Three layers read every message and voice turn: fixed rules, a word list we update without a release, and a check that catches slang and coded words.

  • The crisis reply is fixed and written by people, never by the AI, with the helpline for the student’s country.

  • Your designated contacts get an alert at once, by email and in the Safety console, and can acknowledge, resolve or mark a false alarm.

  • The teacher sees a quiet flag to check in. The alert email never contains the student’s words.

  • Coursework is not an alert: "why does Ophelia die?" is a literature question, and is treated as one.

What this rests on

  • When a student says something that may mean they are in danger, Zelvi stops tutoring, says it is an AI, gives crisis resources for their country, and alerts the institutionIt recognises crisis language in English, Spanish and Hindi, and does not raise an alert on coursework such as "why does Ophelia commit suicide". The alert email never contains the student’s words.
  • Each institution’s data is isolated at the database level, not by application codeThe database itself refuses to return another institution’s rows, whatever the application asks for.
  • Zelvi replies in 22 languages and keeps the course’s English terms beside the translationA question in another language is searched against the course with an English rendering, so the answer can still cite the page.

The rest of the platform