AI Agent Management
Data Sources
Add, train, and maintain the knowledge your agent uses to answer customer questions.
Choose a source
Open Build → Data sources for the selected agent. The overview shows Training data, Active sources, and Workspace vectors, followed by Add a source and the Connected sources list.
| Source | What to add |
|---|---|
| Website | Public website pages and support articles discovered from URLs. |
| Documents | PDF, DOC, DOCX, TXT, or CSV files with readable support content. |
| Plain text | Written instructions, notes, or short knowledge articles. |
| Q&A pairs | Specific questions and their approved answers. |
Add a source
- Choose the source type under Add a source.
- Add the URL, document, text, or question-and-answer content requested by the editor.
- Review the content and start processing.
- Return to Connected sources to check status and published document counts.
- Open the source to inspect its individual pages or content.
Use the source-type filters and search to find an existing source. Training data shows the allowance for your agent; keep the selected content within that allowance.
Manage website pages
Open a website source to see its included URLs, training state, and last sync. Filter URLs by path, filter by status, or sort by name or size. Expand a path group to inspect the pages inside it.
| Control | Use it to |
|---|---|
| Add page | Add another URL to the website source. A page that is already in the source is not added again. |
| View | Inspect or edit the saved content for a page. |
| Re-scrape | Fetch the remote page again and retrain it. This replaces saved page edits, so you are asked to confirm. |
| Include / Exclude | Choose which pages belong in the agent's knowledge. Exclude a page to remove it. Including it again before you train leaves the source unchanged. |
| Settings | Exclude paths, choose the scraping location, turn on slow-scraping, include query parameters, or turn on automatic weekly retraining when your plan includes it. |
| Resync | Fetch every included page again and retrain the source. This replaces saved page edits, so you are asked to confirm. |
| Train all included URLs | Train the included pages from their saved content. On a phone the button reads Train included; it reads Retrain once the source is trained. |
For a batch change, choose Select, tick the relevant URLs, and choose Include selected or Exclude selected. The selection applies to the current view; check it before applying the change.
Understand training status
| Status | Next step |
|---|---|
| Processing / Queued / Training | Let the current job complete and review its progress. |
| Trained | Test a customer question that this page should answer. |
| Changed / Not trained | Review the content and train it before relying on the update. |
| Fetch failed / Not found / Error | Open the reason, check the public URL, and retry after correcting the issue. |
| Excluded | Include the page if the agent should use it. |
| Over limit | Review the training allowance and reduce the selected content or change capacity. |
Check the answer
After training, open Playground and ask the exact question the new content should answer. Compare the reply with the source text. If the answer is incomplete, improve the relevant page or add a focused Q&A and test again.
Keep knowledge current
- Re-scrape a page, or Resync the whole website, after product, policy, or support instructions change.
- Exclude duplicate, irrelevant, or outdated pages.
- Review the reason for unsuccessful pages instead of repeatedly retraining the entire source.
- Retest the same customer questions after important updates.