Capture user feedback (👍/👎)
What it does
After an assistant message appears in chat, you can click 👍 (thumbs-up) or 👎 (thumbs-down) to rate the response. An optional text field expands for free-form critique. The feedback is stored with the completion's provenance and can be aggregated into training datasets.
Prerequisites
- At least one chat message has been sent and received (guide 01).
- The assistant's response card is visible in the chat panel.
Steps
1. Send a chat message
Open the chat panel (⌘K / Ctrl+K) and send a message (guide 01). Wait for the stream to complete.
2. Find the feedback buttons
Hover over the assistant message card (the one with the model's reply). The 👍 and 👎 buttons appear in the top-right corner of the card.
3. Click a rating
Click 👍 for a positive rating or 👎 for a negative rating. An input field expands below the message.
4. (Optional) Add a critique
Type a free-form comment in the text field (up to 4,096 characters). Examples:
- "Model hallucinated; there are only 50 states, not 51."
- "Response was too long; please be more concise."
- "Perfect answer!"
5. Confirm submission
Click Submit or press Enter. A brief toast confirms the feedback was recorded.
(Direct API) Submit feedback via curl
curl -s -X POST \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $BIGWAVE_TOKEN" \
-d '{
"rating": 1,
"label": "good",
"text": "Helpful and accurate response."
}' \
"http://localhost:8080/v1/chat/completions/{completion_id}/feedback" | jqReturns 202 Accepted with the assigned feedback_id.
Verify
- After submitting feedback, the thumbs button is highlighted and the text field closes.
- Run:bash
curl -s -H "Authorization: Bearer $BIGWAVE_TOKEN" \ "http://localhost:8080/admin/feedback?limit=5" | jq '.results[0]' - The feedback row shows
rating: 1(or-1), your text, and the completion's provenance.
Common issues
| Symptom | Cause | Fix |
|---|---|---|
| Thumbs buttons don't appear | Message is still streaming | Wait for the stream to finish before hovering |
| "Request failed (422)" on submit | rating is 0 (neutral) — not yet supported | Use only 1 (up) or -1 (down) |
| "Request failed (404)" on submit | completion_id is unknown or belongs to another workspace | Check that the completion exists in your workspace by listing inferences |
| Feedback appears but text is empty | Optional text field was left blank | This is valid; text is optional (only rating is required) |
Related
- Send a chat message — creates the completions you rate here.
- Inspect inferences (Observability page) — see metadata about rated completions.
- Upload or build a dataset — use feedback to build training datasets.