Skip to main content
Feedback is the end user’s reaction to a specific agent run: a score (thumbs up or thumbs down) and up to 2,000 characters of free-form text. It is tied to a run by ID and available as a filter, dimension, and Insight Search input across the whole product.

​
How feedback is submitted

Feedback is submitted from your application against the run ID you generated when the agent ran.
1

Generate a run ID up front

Pass an explicit tcc.runId when you call your agent. Return it to your client so it can be referenced later.
2

Submit feedback

Call your framework’s feedback helper (for example submitFeedback in the AI SDK integration) with the runId, an optional score, and optional text.Every request must include at least one of score or text.
See each framework integration for the exact syntax.

​
Analyzing feedback

  • Feedback filter on the runs list. Filter to runs with positive or negative feedback, or with text feedback only.
  • Feedback ratio in Recaps. Weekly and monthly Recaps compare positive-to-negative ratios and highlight what negative feedback is about.
  • Per-user and per-org feedback. Once user and organization identity is attached, feedback can be broken down by account or user.
  • Free-form questions in Insight Search. Insight Search can join feedback with tools, models, and metadata.

​
Example questions

Are users generally happy this week? Show the ratio of positive to negative feedback and what the negative feedback is about.
Which tools appear most often in runs that received a thumbs down?
Show me sessions from orgId=acme-corp where the last run had negative feedback.

​
API access

See the API reference.

Traces

Inspect the run behind a piece of feedback.

Patterns

Detect frustration even when the user doesn’t leave explicit feedback.

Insight Search

Cross-slice feedback with tools, models, users, and orgs.

Recaps

Feedback ratios and trends over time.