User Feedback Configuration
GPT-RAG includes a User Feedback Loop feature that lets users evaluate
assistant responses through the UI. In the currently released classic
Container Apps topology and the explicit classic fallback, feedback is
processed by the orchestrator and stored in Cosmos DB. The published
hosted component matrix implements separate
user history, feedback, and delete APIs in UI v2.6.0. Hosted-panel is now an
explicit supported topology in the pinned umbrella matrix, but
PANEL_HISTORY_ENABLED and PANEL_HISTORY_OWNER_BINDING_VALIDATED remain
deployment-published false, so these routes continue to return 503 until
their separate evidence procedure completes.
For the component contract, the UI BFF checks the caller's validated oid
against the owner index before reading managed Conversation messages or
metadata. Missing and non-owner resources both return opaque 404. Feedback
stores bounded, sanitized rating/category/comment/message-reference metadata
only; message bodies and citations remain in managed Conversations. Delete
removes the managed Conversation first and then metadata, returning an explicit
partial result if metadata cleanup fails. Signed pagination cursors are
oid-bound and expiring; tampered, expired, or cross-user cursors return 422.
These routes return 503 while panel/history gates are off.
User feedback stored in Cosmos DB
By default, basic feedback (thumbs up/down) is enabled, while detailed ratings (star rating and comments) are disabled. Administrators control these options through Azure App Configuration.
Feedback Types
When enabled, users can provide star ratings and text comments for richer feedback that captures both satisfaction and reasoning.
User providing rating and comment feedback
Configuration Settings
The behavior of the feedback loop is controlled by key-values in Azure App Configuration:
- ENABLE_USER_FEEDBACK → Default:
trueControls whether the feedback feature is available at all.
Key to enable or disable user feedback globally
- USER_FEEDBACK_RATING → Default:
falseControls whether users can provide detailed feedback with ratings and comments.
Key to enable or disable detailed rating feedback
Default Values
ENABLE_USER_FEEDBACK = trueUSER_FEEDBACK_RATING = false
This means feedback is collected by default, but star ratings and comments must be explicitly enabled by setting USER_FEEDBACK_RATING to true.