Training in Command goes beyond uploading files. Simulations replay scripted conversations against an agent before you ship a change, and Human Feedback turns real conversations users flagged into reviewed, annotated training material.
In this section
- Run scripted simulations to validate agent behavior before release.
- Sync flagged conversations from a connected agent into Human Feedback.
- Read AI review status and scores, and add human reviewer notes.
- Re-pull feedback cleanly with Clear & Resync when data drifts.
What's inside
- 01
Simulations
Open Training → Simulations from the sidebar to run scripted conversations against an agent and validate its behavior before shipping changes.
- 02
Human Feedback overview
Training → Human Feedback lists conversations users flagged for review. Each row shows the agent, conversation title, AI review status, score, and human review notes.
- 03
Syncing feedback
The Sync button pulls conversations from the connected agent. After the first pull it becomes Clear & Resync, which truncates the local copy and pulls fresh. Both open a confirmation modal where you choose the window: All time, or on/after a specific date. If nothing matches, the UI shows a No results notice instead of waiting — completion is detected via an agent metadata stamp, not a fixed timeout. Sync state and last-sync timestamps are stored per agent.
- 04
AI review and human notes
Open a conversation to see the full transcript, the agent's automated review and score, a summary, and a notes field. Notes are stored per row and visible to other org members.
At a glance
| Review status | How it is derived | What it means |
|---|---|---|
| Pending | Queued, not yet started. | Wait or trigger a sync |
| Running | Processing started, no completion recorded. | In flight |
| Completed | A review or completion timestamp exists. | Ready for human notes |
| Failed | The AI review returned an error. | Re-run the review |