Skip to main content
The Request Overview page is the fastest way to debug one specific request and answer:
  • “What happened on this exact input?”
  • “Would a different system prompt have fixed it?”
  • “How would a different model have responded?”

​
How to find it in the Portal

You’ll typically open a request from:
  • an Intent Group’s Requests tab (real traffic), or
  • Test → Evaluations (open a request from a row), or
  • Test Set / Test Run request tables (open a request from a row).

​
What you can do on the Request Overview page

​
Inspect the original request

Review the original request’s:
  • input messages (including system + user messages)
  • assistant response
  • request metadata / IDs (useful for debugging)

​
Create reruns to test prompt + model changes

Use the Versions panel to iterate quickly:
  1. Select Original (baseline).
  2. Click New Rerun to create a new version.
  3. In the rerun version, edit inputs (for example: tweak the system prompt, adjust messages, switch model).
  4. Click Run Request.
This workflow is ideal for comparing “what would have happened” if you changed your messages or model selection for the same request.

​
Evaluate each version with Sentinels

Run Sentinels against the original and any rerun to compare pass/fail outcomes per version:

​
Step-by-step recipe

If you want a guided workflow (rerun → evaluate → optionally add to a Test Set), start here: