Training Your Agent
Make saved sources available for answers. Uploading and training are different steps
Training Your Agent#
Training prepares saved knowledge so the agent can use it. It is not the same as uploading, pasting, or discovering a website.
Open Agent Studio → Data sources. The button is Train agent the first time and Retrain agent after a completed training. Training runs in the background. You can keep working.
Files, text, Q&A, websites, and Google Drive affect answers after training. Structured data, time-sensitive content, and social posts are prepared in their own source workspace. A saved row, a discovered URL, or an extracted file is not proof the agent can answer from it.
What to do#
| Step | What you do | Ready when |
|---|---|---|
| Add the source | Add source, then upload, paste, connect, or select pages | The row shows the content you intended |
| Wait for processing | Leave discovery, extraction, or sync running | The status is no longer processing, unless it needs attention |
| Train | Choose Train agent or Retrain agent when it is offered | Status is Trained, or the source type shows its own ready state |
| Check privately | Ask a question in Playground | The answer matches the source, or honestly says it does not know |
If the button is unavailable, read the reason on it. Typical reasons are that processing is still running, training is already running, or nothing has changed since the last training.
Status you will see#
| Status | Meaning |
|---|---|
| Processing | The source is still being read. Training waits for it |
| Ready to train | Saved and waiting for Train agent |
| Training | Training is in progress |
| Trained | Included for answers |
| Needs attention | Open the source and use the recovery action it offers |
| Paused or Pending setup | Something is still required before training, such as a website sync |
| Removed | Already out of answers. The next training permanently deletes it |
Website counts can show discovered pages that are not downloaded or trained yet. Open the page list before assuming the whole site is ready. Details are in Website sources.
When to retrain#
Retrain after you add or change files, text, Q&A, website pages, or Drive documents that the inventory still marks as needing training. You do not need to retrain one source at a time. Add the related updates, then train once.
You do not retrain just because you changed a greeting or turned on a channel. Do retest in Playground after guidance or tool changes.
Removed sources stop being used before the cleanup training finishes. Do not delete the working copy of a source until its replacement is trained and a private question succeeds.
If training looks stuck#
- Large files and large websites take longer. Refresh and read the status before starting the same work again.
- A failed page does not mean the entire website failed. Retry the page the inventory offers.
- If the answer is still missing after Trained, the fact is not in the source, or the question does not match it. Add the fact or a Q&A pair, retrain, and test the customer's wording.
- Ask in Playground, not on a live channel, while you are checking.