
The message feed
Processed emails are displayed as cards showing:- The email subject line
- The sender’s name and email address
- The category Mary assigned (e.g.,
→ Human Request)
Filtering by category
Use the tab bar at the top to slice the feed by category. Click Unsubscribe to see only unsubscribe replies, Bounce for bounces, and so on. Recents shows the most recently processed emails regardless of category.
Inspecting an individual email
Click any message card to open the detailed view. The detail panel is structured around the full processing pipeline, so you can see exactly what happened at each step.
- The full email body as it arrived
- The classification Mary assigned and her plain-English reasoning for why she made that call
| Section | What it shows |
|---|---|
| Classify Email | Mary’s full reasoning for the category she chose |
| Extract Fields | The values Mary pulled from the email body (if extraction was configured) |
| Fetch Data | Any data retrieved from external systems (e.g., Marketo) before extraction |
| Execute Actions | Which actions ran and their outcomes (success / failure / details) |
| Original Email | The raw email as it arrived, headers and all |
Re-running a misclassified email
If you spot an email that was classified incorrectly, you don’t have to wait for a similar reply to arrive again to test your fix. Click Re-run on the detail view and Mary will reprocess the email using your current category configuration. This is the standard workflow for tightening up a category prompt:Find a misclassified example
Spot it in the feed or via Search.
Adjust the category prompt
Edit the category that should have caught it — or the one that incorrectly caught it.
Lock it in with a test
Add the email to the Test Suite as a regression check so the fix can’t quietly break later.