
The agent answers from the workspace's knowledge and catalog and the Inbox shows the receipt — `Sources retrieved · <sources>` when retrieval found chunks, `⚠ unsourced — review` when it did not.
‘Does it run small?’ — answered from your catalog
The agent reads your catalog and reviewed knowledge before it replies, and the Inbox shows the receipt on every answer. If your knowledge doesn't cover a question, turn on human review and it comes to you.
Start free









