Feedback
Let an end user rate an assistant reply with a thumbs up or down and an optional reason.
Rate an answer
POST /v0/sessions/{sessionId}/messages/{messageId}/feedback
Browser request with a publishable key, an allowed origin and a verified user token. A dev-mode anonymous session, which has no token, is refused with 401 invalid_user_token.
Prop
Type
curl -X POST "$BASEERA_URL/v0/sessions/ses_01JC9X4M2K8Q/messages/msg_a1/feedback" \
-H "Origin: https://app.example.com" \
-H "x-baseera-key: $BASEERA_PUBLISHABLE_KEY" \
-H "x-baseera-user-token: $USER_TOKEN" \
-H "content-type: application/json" \
-d '{ "rating": "down", "reason": "The answer used the old price." }'Response, 200 OK:
{ "rating": "down" }Rules
messageIdmust be an assistant message of that session whose turn has finished. Otherwise:turn_not_found(404), orturn_in_progress(409) while the message is still being written.- The session must be this user's and this agent's, or you get
session_not_found(404). - One rating per turn. Rating again replaces the earlier one, including changing a thumbs up to a thumbs down.
- A
reasonthat breaks the rules isinvalid_request(400), and the message never repeats it. - Anything shaped like a key is replaced before storage. The reason is untrusted text from an end user. It is never given to a model and never logged.
- Limits: 30 ratings a minute per user and 600 per agent (
429 rate_limitedwithRetry-After). Body at most 8 KB.
What happens to a rating
The rating is linked to the turn, so to its trace and record. In the console you can filter the sessions list by thumbs down, and save a thumbs-down turn as an eval test case from the session page.
In @baseera/react, useBaseeraFeedback() returns rate(messageId, rating, reason?) and the current ratings. <BaseeraSurface> shows thumbs under every finished reply.