[ SET THE CONDITIONS FOR CLASSROOM AI ]

Teacher controls

Choose class instructions, model access, shared skills, and the conversations you need to review.

Join your school Talk about your school

NexusIllustrative example

I'm stuck on 2x + 3 = 11. Where do I start?

Nexus

Which operation would undo the +3? Apply the same operation to both sides.

Leaves the next step to the student.

Plan the class policy

Explain when AI is allowed, what students must produce themselves, and how they should acknowledge assistance.

Configure the workspace

Select available models and add class instructions that support the assignment. A guardrail guides behavior, but teachers should still review important work.

Review in context

Teachers can review prompts and answers from school conversations in assigned classes. Use that record to follow up with a student or refine the class instructions.

Set the level of help for the task

Choose hints, guided support, or a fuller answer with critique. Add instructions such as "ask for an attempt first" or "give feedback on the argument without writing the essay." The policy is attached to the class conversation so school work has an explicit context.

Restrict the available models

Start with the models approved by the school. Narrow that set for a class when a lesson needs a particular capability or cost profile. Students cannot authorize a model by asking for it in a prompt; access is checked by the service.

Cited school knowledge

Let students ask about approved school material and open the cited source.

Shared skills

Share a routine students can reuse when practicing or asking for feedback.

Review prompts and all generated responses

The class review view connects a student's question with the answers returned for that conversation, including multi-model responses. Review is scoped to the classes assigned to the teacher. It does not provide access to unrelated personal conversations.

Test rules with the requests students may actually make

Try direct requests for a finished answer, attempts to override the instructions, and ordinary requests for clarification. A useful rule should preserve legitimate learning help while resisting shortcuts. Safety checks reduce risk, but no AI system can promise that every output is correct or appropriate.