Creation and Data Sources

This page follows a timeline: first find an existing knowledge base in the list, then create a new knowledge base and connect its initial data sources. After creation, data sources will continue to be added and removed and require routine maintenance. Finally, confirm who uses it—changes to a knowledge base affect every agent bound to it, so it is worth reviewing the impact before making changes.

Corresponding pages in the product:

Entry point

Description

Knowledge base list (Knowledge Bases tab)

Search, create, edit metadata, delete, and open advanced configuration or chat

Advanced Configuration · Data Sources

List associated files / tables, add data, enable or disable, delete, and more

Advanced Configuration · Agent Associations

List agents explicitly bound to this knowledge base

Finding a Knowledge Base

Knowledge bases are displayed as a card list showing the knowledge bases you are authorized to access in the current workspace. You can search by name or description, and each card shows scale information such as the number of files and tables.

There are only two basic information fields: the name should directly reflect the business domain or scenario, such as “Sales Operations Knowledge Base”; the description should explain its purpose and the scope of questions it can answer, making it easier for agents and people to identify. Keep the name and description consistent between the list and details after editing them. You can see only authorized knowledge bases; see User Permissions for authorization rules.

Common card operations:

  • Edit metadata: Change the name, description, and other information

  • Edit source data / Advanced Configuration: Open data source, semantic, and other maintenance settings

  • Chat: Create or open a chat session based on the knowledge base

  • Delete: Delete the entire knowledge base (different from deleting a single data source)

Creating a Knowledge Base

  1. Click Create Knowledge Base in the list.

  2. Select a data source method (choose one of two options in the creation wizard or complete the relevant steps):

    • Local upload

    • Select from Catalog

  3. Enter basic information such as the name and description.

  4. Complete data selection as directed by the wizard (and the indexing-related configuration required at creation time; see below).

  5. Confirm creation. The creation API creates a knowledge base with initial data sources. After creation succeeds, the materials enter the processing workflow; successful upload / association does not mean they are immediately available for Q&A.

Index Configuration

When a knowledge base is created, the system binds index configuration related to text vectors (called “Knowledge Base Index” in the UI). You must select a text Embedding model (which may include the system default). You can optionally enable image indexing and select an image Embedding model; when enabled, it is used for image-related retrieval.

Note The available model list is determined by the platform model library. Index configuration is bound during creation, and the document parsing workflow writes to the corresponding index.

Two Ingestion Methods

Files added through local upload are written to the knowledge base’s own storage space. For files or tables selected from Catalog, the original data remains in its original Catalog location, and the knowledge base records only the association. This is why removing a Catalog source from a knowledge base does not delete the original file in Catalog (see below). The same Catalog data can be referenced by multiple knowledge bases. To decide which method to use, first check whether the data is already in Catalog: if it is, select it directly from Catalog; for material on your local machine that has not been uploaded, use local upload.

Local upload adds files from your local machine to the knowledge base. The wizard distinguishes them by ingestion type:

Ingestion type

Applicable files

Result

Unstructured

Documents, audio, video, images, and so on

Enter the parsing workflow for retrieval and Q&A

Structured

CSV, XLS, XLSX

Import into knowledge-base-side structured tables according to table definitions (with field preview and configuration for Sheet, header row, primary key conflict policy, and so on)

You can select a single file or a folder. To add data from an external system to a knowledge base: direct connection to an external database directory is not currently supported. Export the data as files and upload them locally, or first load them into Catalog through Data Ingestion, and then use “Select from Catalog.”

Select from Catalog lets you choose document files or data tables from the Catalog tree and add them to the knowledge base (the UI indicates that only document files or data tables are supported). You can view a summary of the number of selected files / tables:

Type

Description

Catalog file

Reuse a file already in the workspace; the knowledge base manages the association, while the original file remains in Catalog

Catalog data table

Add it to the knowledge base as a structured source for semantic configuration and NL2SQL

Removing a data source from a knowledge base does not delete the original file or business table in Catalog. The knowledge base removes only the association, not the data itself.

Validation After Creation

Confirm that the data source processing status is Completed (or that a table source is available), and then validate it in Chat and Retrieval. If structured tables are present, also configure Segments and Semantics.

Ongoing Data Source Maintenance

Creating a knowledge base is not the end. As the business changes, materials are continually added or retired. On the Data Sources tab in Advanced Configuration, you can continue to add data (appending local or Catalog sources). The list shows associated files and tables by their degree of structure and summarizes the number of unstructured files and structured tables:

Column

Description

Name

File / table name

Type

File, table (including assets such as volumes)

Size / rows

File size or number of table rows

Catalog path

Location in Catalog

Processing status

Completed / Processing / Failed / Pending ingestion, and so on

Updated at

Most recent update

Enabled status

Whether it participates in retrieval (including expiration-related states)

Operations

Details, download, validity period, delete, and so on

For the meaning of processing statuses and for enablement, validity periods, tags, and versions, see Data Content Management.

Difference Between Deleting a Data Source and Deleting a Knowledge Base

These actions have completely different scopes of impact. Distinguish them before proceeding:

Operation

Impact

Delete a single data source

Removes only the association between that material and the current knowledge base

Delete the entire knowledge base

Affects everyone who can access the knowledge base and all bindings; confirm before proceeding

Agent Associations

Changing data sources, segments, or semantic configuration affects the answers from all chats and agents bound to the knowledge base. Before making changes, confirm the extent of the impact instead of discovering afterward that the answers from an agent in use have changed.

The Agent Associations tab in Advanced Configuration lists agents explicitly bound to this knowledge base ID (including name, description, status, number of bound knowledge bases, update time, and so on), and you can navigate to the agent side to inspect them:

  • An empty list means that no agents are currently explicitly bound to this knowledge base.

  • Binding / unbinding a knowledge base to an agent is done in Agents; this page is used to view the impact from the knowledge base side.