Data Sources

Data sources are where you provide the system access to your data. When you load any text information into the system, it reads the text, splits it into chunks, and uploads in a special storage (currently, we user a vector database). When a user query comes, we looks in the datastore to find the most fitting chunks, and then pass them to the ChatGPT engine which looks for an answer and build a human-like response.

Add a data source

As soon as a project is created you may add the data source. A data source is a piece of information to be indexed and searched by our engine. Read more on the data source documentation page (opens in a new tab).

Click the "Add a new source" button and then select the corresponding type of data sources:

Let's review all these datasource types.

Plain text

After you create a new text data source, you will be redirected to its page:

To save the data source under a specific title, enter it and click the Save changes button to save.

The plain text data source can contain multiple pages of text. To add a new page of text, click the Add text content button. In the modal window, enter the title and the content of a data source:

Then click the Save & Index button to save the changes and immediately index the content in our database to by used later by the chatbot.

PDF files

When you select to create a file-based data source, just click the Select or drag file(s), select your file, then click the Upload button.

The process of uploading followed by indexing will start. You will something like that while it's in progress:

When it's done you can click on a page to review and edit the text content:

Web site

You may enter the full URL, or a subfolder, or even specify pages manually. Another option is to use the sitemap file. Let's review all these options:

Let's review all the options for finding web pages.

All pages within a domain (automatically)

When this option is selected, the crawler will look for any links, put them into a special list and then load the pages to find new links. The crawler is optimized in a way to avoid multiple hitting the same pages. As soon as a page is loaded, the crawler saves its content in a special in-memory database. This data will stored there for 7 days. So, if you'd like to just re-index the content of your website for some reason (without re-loading the pages), it will take much less time.

Only within a specific path (like inside of a folder), automatically

You enter a path you want all the pages under would be indexed:

Read pages from the site map

You can enter the exact URL of your website's sitemap file, or even just your website's URL and our system will find a sitemap on its own:

Manually entered pages

Also, you can enter URLs manually and then click the plus button to add them to the list:

You can do it any moment but before you start indexing.

Use the Crisp conversations as data sources

If you have installed the Crisp plugin, you can easily add the conversations as data sources. In the Crisp dashboard, find the Enum widget (it's called Support AI Chatbot):

Click the section Train the chatbot with this conversation, then click the Add this conversation button. Then the Crisp modal window is open where you can delete messages that you don't want to be indexed (or have some private information) and edit them:

After preparing the conversation, enter the data source title, and click the Add this conversation button in the modal. The process of indexing may take up to several minutes (usually, it's just seconds) and you will see the corresponding message:

If you click the link you will be redirected to your dashboard where you can review and edit the newly created data source.

Indexing webpages

After you find or add the pages to be indexed, select them and then click the Start indexing button. The stats of indexing pages will be updated automatically. All indexed pages will be unchecked.

You can add as many pages as you want within your limits. After the indexing is done you will see the updated numbers and icons:

Indexing parameters

Density of data

You can specify if you want data to be stored densely, or sparced. Sometimes, when parsing a website, our system leaves a lot of empty space between text data (that looks like empty lines). Usually, it doesn't affect to the chatbot's quality but sometimes, removing these gaps could improve it significantly.

Include links

This option allows you to include links in your extracted text alongside the other text. For example, if your data source has the following HTML code ...in our <a href="https://mywebsite.com/documentation.html">documentation</a> the full link will be included as "...in our https://mywebsite.com/documentation.html (opens in a new tab) documentation" if the option is selected. Even though it may provide more valuable information, in some cases it may worsen the results, so always test this option with indexing just a couple of pages with the following testing the chatbot on the Interaction page to make sure the chatbot quality is appropriate.

Force reload pages If switched ON, the content of the selected pages will be read again. If switched OFF, the content will be obtained from the memory (if pages already have been read) which can dramatically speed the indexing process.

Re-indexing website

You may completely re-index your website, make crawling from scratch, and then select and re-index pages, or you can change and re-index an individual page as it's done with any other type of data source.

To re-index the whole website, just select the pages to be indexed and click the Start indexing button. To re-index the individual page, find it and click the button to edit (not necessarily) and index:

Ignored pages

You may want to mark some pages as ignored. It can be done at any moment. The ignored pages will not be indexed or crawled. To mark a page as ignored, just click on the button in the corresponding column:

Deleting data

You can completely delete the data obtained from some pages, from the index database. To do so, just select pages, then click the Remove page data button. You will have to confirm your intent to delete this data.

To remove this flag, click on the same button again.

Filtering, searching, and manipulation with pages

Filter page by any part of the URL

Just start printing in the Filter field and you will see how the list changes:

https://github.com/user-attachments/assets/d986ec7e-954b-46b0-931d-7b37c5bfe29a (opens in a new tab)

Then, you can apply an action to the selected pages:

Here is the list of available actions:

• Ignore (mark pages to be ignored during indexing)
• Unignore (unmark pages to be ignored)
• Remove (remove from the list and the corresponding data from the indexing database, so the content will not be available for your chatbot anymore)
• Remove indexed content (remove the content only, but leave the page in the list)
• Sync index status (syncing the index status of content with the status in the list - you only need this option if something went wrong and your list is not updated properly)
• Unselect (remove selection).