Privacy Notice

How Munkey AI handles school data

Munkey AI is designed for institution-managed course support. This notice describes the data the app handles, where it is processed, and the controls reflected in this repository. It is not a legal contract and should be paired with institution-specific policies and vendor agreements.

Data Collected

Account, membership, course chat, and uploaded course material data needed to deliver the product.

Institution Context

Access is scoped by institution and course membership before chat, materials, and management features are exposed.

Current Posture

This repository removes unsupported compliance claims and adds transparency and private image handling.

Data categories used in the app

`Account data`: email address, authentication identifiers, and institution membership.

`Course data`: course rosters, roles, and uploaded PDF materials.

`Chat data`: conversations, citations, structured response forms, optional prompt images processed in memory during a request, message-linked thumbs ratings with fixed-choice reasons, and aggregate timing measurements.

`Quality and performance data`: turn and course identifiers, route and model versions, restriction and feature flags, retrieval counts, token counts, stage timings, outcomes, and stable error codes. These records do not contain prompt, response, or other free-text copies.

Third-party processors used by the product

`Supabase` provides authentication, database, and storage infrastructure.

`OpenAI` processes prompts, course context, and embeddings to generate or ground assistant responses.

`Pinecone` stores vector embeddings derived from uploaded course materials for retrieval.

`LlamaParse by LlamaIndex` processes uploaded PDFs only when advanced layout analysis or OCR is requested or normal text extraction is not reliable.

`Vercel Workflow` runs durable background processing for uploaded course materials and records operational workflow state.

Institutions should review and execute their own vendor terms, direct-control language, retention commitments, and approved-use policies before deploying the system.

User Transparency

Signed-in users can review their account identifiers and current account context from Account Settings.

Assistant-message ratings are linked to the rated message so the course can measure answer quality. Feedback reasons are fixed choices; the feedback form does not collect free text.

Institution admins can review institution details and manage users from the protected administration views.