Skip to main content

Understanding Knowledge Stores and Narratives

A Knowledge Store is a collection of information that an assistant can reference when answering questions. Knowledge Stores let you group information for a specific use case and control which assistants and users can access it.

An assistant can search the information available to it through its configured Knowledge Stores. It does not automatically have access to all data in the Data Studio database.

Narratives populate Knowledge Stores with searchable content. A Narrative reads data from a source in Data Studio and converts it into text vectors—numerical representations of text that support searches based on meaning. Each Knowledge Store stores these vectors in an underlying vector table

A Knowledge Store can receive data from one or more Narratives.

Types of Knowledge Stores

There are three types of Knowledge Stores in Data Studio: Document, SQL and System.

  • Document Knowledge Stores contain unstructured data added from Document Stores through Document Narratives.

  • SQL Knowledge Stores contain structured data added from local database tables through SQL Narratives.

  • System Knowledge Stores enable out–of–the–box assistant capabilities (see Default Available Agents for a full list).

Organizing Knowledge Stores

The way you organize Knowledge Stores directly affects retrieval accuracy, maintainability, and reuse. Because a single Knowledge Store may be referenced by multiple assistants, you should clearly define its scope and content. Use the following best practices when creating Knowledge Stores:

  • Create a separate Knowledge Store for each distinct use case or business function. Clear boundaries between Knowledge Stores make it easier to manage content and ensure that assistants retrieve only relevant information.

  • Keep data domains separate. Avoid combining unrelated data into a single Knowledge Store, as this can introduce noise into retrieval and reduce the accuracy of responses.

  • In general, prefer smaller, well-scoped Knowledge Stores that support a specific use case

Understanding the Knowledge Store Interface

Navigate to Management > AI Assistants > Knowledge Stores, and you will see a table of your existing Knowledge Stores. The table has three tabs: System, containing Knowledge Stores that Data Studio has automatically created to support built-in agents; SQL, containing Knowledge Stores designed for structured data from tables; and Document, containing Knowledge Stores designed for unstructured data. The details contained in each tab are as follows:

Name

The name of the Knowledge Store.

Status

Indicates the current operational state of the Knowledge Store:

  • Pending Creation—The vector table has not yet been created.

  • Creating—The vector table is being created.

  • Ready—The vector table has been created successfully.

  • Error—An error has occurred during the creation of the Knowledge Store. Disable and reenable the Knowledge Store to restart the creation process.

  • Disabled—Search is disabled for the Knowledge Store.

Knowledge Store status does not reflect the status of any Narratives associated with it.

Search Enabled

Indicates whether assistants can search the Knowledge Store. When search is disabled, the Knowledge Store retains its data, but assistants cannot use that data when answering questions, even if the Knowledge Store is assigned to the assistant.

Description

Describes the Knowledge Store’s content and intended use. The assistant uses this description to determine whether to search the Knowledge Store when answering a question. Describe the scope of the information and the kinds of questions or tasks it can support, including relevant context such as subject matter, content types, or time periods.

You can expand a Knowledge Store by clicking the + icon to view the following details about the Narratives associated with that Knowledge Store:

Name

The name of the Narrative.

Status

Indicates the current operational state of the Narrative:

  • Loading— The Narrative is processing data that existed in the source when the Narrative was created.

  • Processing— The Narrative is processing data added to the source after the Narrative was created.

  • Ready— The Narrative has been created successfully and is monitoring for new or changed has to ingest.

  • Disabled—Search is disabled for the Narrative.

  • Error— An error has occurred. This state is only shown once the Narrative reaches the maximum number of failures as defined during Narrative creation. Until then, the Narrative may be in the Ready state.

Total Items

The total number of items currently stored in the Knowledge Store.

Pending Items

The number of items that are waiting to be processed and added to the Knowledge Store.

Failed Items

The number of items that failed during processing. Click this field to show the error details, Narrative Template and the impacted document or SQL statement.

Items up-to-date

Indicates what percentage of items in the Knowledge Store have been processed and are ready for search.

Time Remaining

Provides an estimate of how long it will take to finish processing any remaining items.

Filtering the Display

To filter the display, click the + icon to use the drop-down menus to filter for certain fields or look for keywords using the type-in boxes above the table.

Each individual field can also be filtered, either in ascending or descending order, by clicking the sort generated description: icon sort.svg icon.

FeedbackOpens in a new tab