How to setup the retraining

The re-training feature provides an automated way to update training by:

Re-reading the specified URL to get the latest content
Processing the updated content to refresh the training data

This enables seamless updates when the source material changes without requiring manual intervention. The system will:

Detect changes in the source content
Update the training dataset accordingly
Re-run the training pipeline
Apply the new training results

This feature is particularly useful for:

Keeping training data current with evolving documentation
Maintaining accuracy as source content is updated
Automating the maintenance of trained models

Selecting addresses for re-training

Retraining single address

In the Training section for website sources click on the calendar icon:

Retraining group of addresses

Select group of addresses by checking the checkbox next to selected addresses and then for the selected addresses you can set up the retraining by clicking on the Retrain button in the header:

After clicking the calendar icon the re-training dialog opens:

Re-training options

You have following options available:

**Retrain now - **to run the retraining right away
Schedule the re-training
- **everyday **(available for Premium subscriptions) - you can select time of the day
- once a week - you can select day of week and time of the day
- once a month - you can select day and time of the day
Select behaviour what should happen if the page is not found anymore on your server. You have following options available:
- DELETE ** **- the page will be removed from the chatbot database
- IGNORE - the page will continue exist in the chatbot database and the chatbot will continue to use the contents from this page

Re-training limits

The system has a limit of 50,000 automatic retrains per month. Each retrain operation counts as one URL being scraped and its contents being updated in the database. You can purchase packages of additional retrains if needed.