AI and data transparency
This page sets out which of the app's skills use AI, what data is sent and which data never leaves your computer.
Updated 15 Oct 2026
- 01
Where AI is used
In the current version, AI fills in structured content for test questions, answer verification, the lesson plan body and its review, slide content, slide theme suggestions and the chat assistant. Everything else runs on your computer.
- 02
Model tiers
Each skill runs on the fast, standard or strong tier; Agentra maps each tier to a specific model. In the Export step of a test, lesson plan or slide deck, a note lists the AI steps that ran, the tier, the actual cost and when the version was made.
- 03
Data sent out
Each call sends only the content the skill needs, under a strict data contract. The assistant sends a structured summary of the open screen, the document passages in the Viewing chips and the latest messages of the conversation, never the original files.
- 04
Data that stays on your computer
The project store, chat history and activation secrets stay on your computer. Imported files are read locally and the original files are never sent; scanned pages and photos are not read into text yet.
- 05
The server stores no content
The server relays content to the model and records token counts, amounts and hashes. Content is neither stored nor written to logs.
- 06
The machine suggests, people decide
AI content carries the Suggestion label until you approve it. There is no button to skip a quality gate; the assistant does not approve for you.
- 07
Three tiers of conclusions
Tier 1 (deterministic) is a sentence the app itself writes for one of its facts, with a Basis line. Tier 2 opens with “Not found yet” for what the assistant has not seen. Tier 3 carries the label “Suggestion, needs human review”.
- 08
AI off per project
Switch AI off in the project details and the app sends no AI calls for that project, including the assistant.