A knowledge base contains information that LLM models can use as grounding or additional context. For example, you could create a knowledge base containing your product support content to help detect inaccuracies in any generated content about your products.
You need an administrator or publisher role on your team to delete knowledge bases. See Knowledge base and data source permissions.
You can create vector store or Kendra store knowledge bases:
- Vector store knowledge bases allow you to maintain your knowledge base files within the inGenious platform.
- Kendra store knowledge bases connect your chatbot to an Amazon Web Services Kendra store knowledge base. You'll maintain your files in Amazon Web Services and sync any changes to your chatbot. You can configure a schedule to sync automatically.
Most of the time, a vector store knowledge base is sufficient for what you need. We recommend contacting inGenious AI support if you're considering connecting a Kendra knowledge base.
Each vector store knowledge base contains up to five data sources. Data sources are created by uploading files, connecting published chatbot content, or crawling the pages of a web site.
The data must be ingested (processed so it can be understood by an LLM model) before it can be used, which occurs when you sync or publish changes to a data source.
Syncing a data source makes the updated content available to TestBot for you to test. Changes to your data source content can impact the response of any LLM models or chatbot functions that use it, so you should sync a data source to test it thoroughly before you publish any changes. You can see the status and type of each data source and the date it was last synchronised on the knowledge bases page.
Knowledge base data is not stored in versions
Changes to the data sources in your knowledge base must be published so that you have control over when your chatbot uses the updated data, but the data is not stored in your chatbot version. If you delete a data source or knowledge base, it is deleted immediately from your live chatbot and can't be recovered. Any LLM tasks or chatbot features that were using the knowledge base will immediately stop working.
To use vector store knowledge bases:
- Create a vector store knowledge base
- Create a data source
- Synchronise each data source
- Publish the synchronised data sources to your live chatbot
You can also:
- Rename a data source or view its history
- Delete a data source
- Rename a knowledge base
- Delete a knowledge base
Knowledge base and data source permissions
| Reviewer | Editor | Publisher | Admin | |
|---|---|---|---|---|
| Create a knowledge base | - | ✓ | ✓ | ✓ |
| Sync a knowledge base or edit a sync schedule | - | ✓ | ✓ | ✓ |
| Delete a knowledge base | - | - | ✓ | ✓ |
| Create a data source | - | ✓ | ✓ | ✓ |
| Edit data source files or urls | - | ✓ | ✓ | ✓ |
| View a data source history | - | ✓ | ✓ | ✓ |
| Sync a data source | - | ✓ | ✓ | ✓ |
| Delete a data source | - | - | ✓ | ✓ |