Managing feedback

Feedback view on the Agents page for reviewing agent traces, annotating agent runs, and promoting high-quality feedback to episodic memory.

The Feedback tab provides a centralized view for reviewing agent traces and capturing feedback that helps improve agent reliability over time. It connects runtime analysis with feedback-driven episodic memory, enabling agents to gradually refine their decision-making without constant prompt rewrites.

Feedback serves two complementary roles:

• Runtime analysis and diagnostics: Review completed agent runs to understand behavior, identify issues, and annotate specific decisions or outputs.
• Input to episodic memory: High-quality feedback can be promoted to episodic memory, allowing the agent to repeat successful behaviors and avoid incorrect ones in future runs.

Feedback overview

The top section of the page provides aggregated insights:

• Agents with most feedback shows which agents receive the highest volume of positive and negative feedback.
• Feedback over time visualizes feedback trends, helping you correlate behavior changes with improvements or regressions.
• Time range filters let you narrow the analysis to recent or historical runs.

Figure 3. The Feedback tab in Agents Instance Management

Adding feedback to agent runs

To provide feedback:

1. Select Add feedback from the Feedback page.
2. Choose an agent runtime job from the list of executed jobs.
3. Review the trace and attach feedback to the relevant span.