Data Content Management¶

Content in a knowledge base is first distinguished by data form, rather than understood through the two UI labels “file / table”:

Form	What it is	Typical sources	How the system uses it
Unstructured data	Documents, presentations, images, audio, video, and other materials	Locally uploaded unstructured files; document files in Catalog	Parse → segment → index → retrieval and Q&A
Structured data	Business tables and metric data with row and column semantics	Catalog data tables; tables created by structured ingestion of local CSV / XLS / XLSX files	Add to the queryable scope + semantic configuration → data Q&A / NL2SQL

Both can coexist in the same knowledge base, but their processing methods and subsequent maintenance priorities differ. The data source list in the UI has type columns such as “File” and “Table.” These describe the carrier (what a data source looks like), not the product’s conceptual division:

A “file” data source follows the unstructured processing pipeline.
A “table” data source follows the structured usage pipeline.
For locally uploaded Excel files and similar content: if they undergo unstructured parsing, they are understood as documents; if they undergo structured ingestion, they are imported as knowledge-base-side tables and then used as structured data.

This page explains how the two types of content are processed after ingestion, how to confirm that they have taken effect, and how to enable or disable them and manage validity periods, tags, versions, and other operations. Corresponding UI: Advanced Configuration · Data Sources list, and document details / table details.

Processing Status and Conditions for Taking Effect¶

After material is associated with a knowledge base in Creation and Data Sources, the file or table appears in the data source list and automatically enters the processing workflow—but a successful upload or association only means that the material has entered the knowledge base; it does not mean that the content can already participate in Q&A. Content participates in retrieval and answers only after processing is complete. For newly ingested material, wait for processing to finish before validating Q&A results. If processing fails, investigate the three areas indicated by the page: file format, file content, or environment configuration.

Processing status	Meaning	Next step
Processing	Parsing, segmentation, or indexing is in progress (mostly for unstructured data)	Wait for completion; questions asked at this point are unlikely to retrieve this content
Completed	The content has entered the retrievable / queryable scope	Validate the results in chat
Failed	Parsing or indexing failed	Investigate the file format, file content, or environment configuration, and re-upload if necessary
Pending ingestion	A legacy source or a source not fully incorporated into governance	Complete the missing information before it can be included in retrieval

To confirm whether material has actually taken effect, check both the processing status and actual retrieval results, rather than looking only at whether the original file still exists—a file being “present” does not mean its content is “usable.”

Unstructured Data¶

After unstructured material is ingested, the system automatically performs preset content processing; users do not need to orchestrate processing tasks manually. The goal is to turn the original material into retrievable, citable content while preserving it for traceability.

Processing Methods¶

Material type	Processing focus	Availability depends on
Documents (PDF, Word, Markdown, web pages, email, and so on)	Content parsing	Whether text can be understood and cited correctly
Presentations, spreadsheet files parsed as documents, and so on	Content parsing	Layout and extraction quality
Images	Text extraction (OCR) and visual information	Clarity and description quality
Audio / video	Transcription before unified processing	Transcription quality

All processing output enters the subsequent retrieval and Q&A flow. Documents emphasize body text; images can rely on text or visual semantics; audio and video depend heavily on transcription.

Storage in the Knowledge Base¶

Each knowledge base has a dedicated system-maintained space in Catalog. Original material and processed output are stored separately to avoid conflating “uploaded” with “effective”:

Knowledge-base directory
└─ <Knowledge-base name>
   ├─ raw_*        Original material (can be divided into volumes by type, such as documents / images / audio and video)
   └─ processed    Processed output for retrieval and Q&A

The directory and workflow are maintained by the system; users do not need to create these volumes manually. Removing an unstructured source from a knowledge base does not delete the original file in Catalog (for Catalog ingestion scenarios).

Viewing in the UI (Document Details)¶

When you open the details of an unstructured data source, common capabilities include:

Area	Content
Preview	Original text / parsed content (depending on the format and parsing results)
Information and governance	Tags, validity period, enablement, force enable after expiration, version history
Intelligent segmentation	Maintenance of retrieval units; see Segments and Semantics

Changes to segments and indexes create version history entries (such as initial import, segment editing, re-embedding, and workflow parsing). You can browse historical versions and set one as the current version. A version is a capability at the individual data source level, not a release version of the entire knowledge base. Segments can be edited only in the currently active version.

Structured Data¶

Structured data is not processed through the same “original text → segmentation → vector retrieval” flow. Instead, it enters the knowledge base as a queryable data scope.

Processing and Usage¶

Key point	Description
Ingestion result	The table enters the knowledge base’s data scope and can be referenced by NL2SQL / data Q&A
Difference from unstructured data	It does not depend on the same raw/processed document segmentation pipeline
What else is needed for accurate answers	Semantic configuration (field meanings, metric definitions, table relationships, business rules, and so on)

Simply adding a table to a knowledge base only defines “which data can be queried.” When business definitions are complex, you must add semantic entries in Segments and Semantics for results to better meet expectations.

Sources include:

Selecting an existing data table from Catalog (removing the knowledge base association does not delete the original Catalog table).
Local CSV / XLS / XLSX structured ingestion (imports the data as a knowledge-base-side table, with configuration for table name, column definitions, primary key conflict policy, and so on; see the creation wizard).

Viewing in the UI (Table Details)¶

Tab	Content
Fields	Column name, data type, whether it is a primary key, default value, description
Statistics	Column-level statistics and more
CREATE TABLE SQL	Table schema SQL
Sample data	Sample rows

Routine Maintenance for Both Data Types¶

Completing processing is only the baseline, not the end. Data sources require continuous governance after they go live—whether they should continue participating in retrieval, whether they have expired, and whether they need tags for team management are separate from whether the content is correct. Processing status governs the latter; the following operations govern the former. The data source list summarizes unstructured and structured entries (often labeled “number of files / number of tables” in the UI). The maintenance operations below apply to data sources in the list generally (document details often provide more complete options):

Enabling and Disabling¶

Control whether the data source participates in retrieval / Q&A. You can temporarily take it offline without deleting it.

Status	Meaning
Enabled	Can participate
Disabled	Does not participate
Expired	Has passed its validity period and does not participate by default
Force enabled after expiration	Has expired but is still forced to participate (requires enabling the “Force enable after expiration” option)

Validity Period¶

After an expiration time is set, the source is automatically disabled from the available scope when the time passes. Changing it to a future date restores it. The validity period only affects enabled status and does not rewrite the content itself. It is suitable for time-sensitive policies, campaign rules, quarterly materials, and similar content.

Tags¶

Used for categorization, filtering, and team governance; tags are not body content.

Download and Delete¶

Download: Export the source file or table-related data for verification, archiving, or external sharing.
Delete data source: Remove its association with the current knowledge base; the original Catalog item is retained. After deletion, agents bound to the knowledge base also lose this source.