menu_bookChapters · Chapter 5 Training dataexpand_more

Chapter 5

Training data

· Published by Osentio

An AI agent only answers from the training data you give it. The more complete the content and the closer it is to what customers really ask, the more accurate the answers. Select the agent, then open the Training Data tab.

Training Data page
Training Data page (Website tab)
No.Description
1The four kinds of sources: Website, Files, Text and Q&A Pairs.
2 3Enter a URL and click Fetch links to add a website source.
4Total size: storage used and your plan's limit. Retraining is not possible above the limit.
5Trainings this month: trainings used and available this month, with the reset time below.
6Retrain agent: can be clicked only when the data has changed.

Website

There are three ways to add links:

  • Crawl links: enter the site's home page; Osentio follows the links within the site to find other pages.
  • Sitemap: enter the address of a sitemap to add all the pages it lists.
  • Individual link: add one specific page only.

Open Advanced options to include only, or exclude, certain paths. When fetching finishes, the links are listed under Link sources, where you can tick and remove pages you do not need.

Link sources
Link sources: expand a source to see how much text was read from each page

The number under each link is the amount of text read from that page. A value of 0 means no readable text was found, so the agent learns nothing from it; add the content as text or Q&A pairs instead. When your website changes, click ⋯ next to the source, choose Re-crawl, then retrain.

Files

Uploading a file
Files tab
  1. Switch to the Files tab.
  2. Drag files onto the dashed box 1, or click inside it to select files.
  3. When the upload finishes, the file is listed under File sources 2 with the upload time and the amount of text read. Right after uploading it briefly shows as processing.
TipSupported formats are pdf, doc, docx and txt. The maximum size of one file is 10 MB on Pro and 20 MB on Enterprise. A PDF must contain selectable text; a PDF made of scanned images cannot be read. The same file cannot be uploaded twice.

Text and Q&A pairs

Adding a text snippet
Text tab
Adding a Q&A pair
Q&A Pairs tab

Text is for information that is not on your website but customers often ask about, such as opening hours, shipping fees or return rules. Enter a Title 1 and the content 2, then click Add text snippet 3.

Q&A pairs are for questions that need a fixed answer. Enter a Title 1, the Question 2 and the Answer 3, then click Add Q&A 4. Use the plus button next to the question to add other ways of asking the same thing.

Retraining

ImportantAfter adding, changing or deleting any training data, you must click Retrain agent. Only then does the agent answer from the latest content.
  1. Check on the right that Total size is within the limit and that Trainings this month has trainings left.
  2. Click Retrain agent and confirm the message that retraining has started.
  3. The page shows “Knowledge base is updating”; the notice disappears by itself when training is done.
TipTrainings are counted per month and shared by all agents under the same workspace creator. Prepare everything you want to add first, then train once.