Document management and parsing

Upload documents, view parsing results, and manage chunks.

Documents serve as the content source for your knowledge base. After uploading, the system automatically performs parsing, chunking, vectorization, and index building.

Importing data

1. Navigate to the knowledge base details page: Click the target knowledge base name on the Knowledge Management page.

2. Click Upload Data: Click Upload Data above the document list to begin the import process.

3. Select a data source: Select a dataset (the default is "Default File Dataset"), then choose your data source:

   Data Source	Description
   Upload Files (default)	Upload local files via the web interface; up to 50 files per batch. Supported formats and size limits are listed in Supported formats.
   Select Category	Choose a file category from the dataset; imports all files under that category and builds indexes. Supports automatic synchronization.
   Select Files	Manually select individual files from within a dataset category; up to 50 files per batch.

   Additional configuration options:

   • Category Configuration (required): Select the category to which the files belong, or click Add Category to create a new one.
   • Add Tags (optional): Assign tags to the imported files. Each tag may contain up to 32 characters.

4. Configure index settings: Click Next, then select a chunking method and segmentation parameters.

   Chunking Method	Description
   Smart Chunking (default)	A robust general-purpose method optimized for most document types, delivering strong retrieval performance across diverse content.
   By Length	Suitable for scenarios with strict token count requirements.
   By Page	Ideal for documents where each page conveys an independent topic.
   By Heading	Best for documents structured with clear headings demarcating distinct topics.
   By Regular Expression	Splits text using a custom regular expression.
   By Delimiter	Splits text based on specific delimiters present in the document.

   Parameter	Default Value	Range	Description
   Maximum Chunk Length	600	10–6000	Maximum number of tokens per chunk.
   Metadata Extraction	Off	On/Off	Automatically extracts document metadata.
   Excel Header Concatenation	Off	On/Off	Appends Excel header information into chunk content.

5. Complete the import: Click Finish. The system automatically executes parsing, chunking, and index building. Once the document status changes to Parsing Completed, it becomes searchable.

Supported formats

Type	Extensions	Size Limits	Notes
Documents	.doc, .docx, .ppt, .pptx, .pdf	≤ 150 MB or ≤ 1000 pages	Electronic PDFs are parsed directly; scanned PDFs undergo automatic OCR.
Spreadsheets	.xls, .xlsx	Recommended ≤ 10 MB and ≤ 100,000 rows	Chunks are generated row-by-row; first row is treated as column headers.
Images	.png, .jpg, .jpeg, .bmp, .gif	Single image ≤ 20 MB; shortest side > 15 px; longest side < 8192 px; aspect ratio < 50	Used for mixed text-and-image scenarios.
Plain Text	.md, .txt, .html	Recommended ≤ 10 MB	Markdown files are chunked by heading; HTML content automatically strips <script> and <style> elements.

Document status

Status	Meaning	Recommended Action
Uploading	File transfer in progress	Wait
Parsing	Extracting structure & text, chunking, generating vectors	Wait
Parsing Completed	Index built successfully; document is searchable	No action required
Failed	Parsing or indexing failed	Review error details and re-upload after correction

Chunk management

Navigate to the knowledge base details page and switch to the Chunk Details tab to view and manage all chunks. Each chunk displays:

• ID and character count
• Preview of chunk content
• Associated document
• Enable/Disable toggle

You can filter chunks by document using the All Files dropdown, or search by keyword using the search bar.

Creating a chunk

Click Create Chunk, then fill in the following fields as needed:

Field	Constraints	Description
Associated Document	Required	Select from already-uploaded documents.
Chunk Title	0–50 characters	Displayed in the chunk list.
Image	Supports .png, .jpg, .bmp, .gif; ≤ 20 MB per image	Optional; used for mixed text-and-image chunks.
Chunk Content	Required; 0–6000 characters	Main body of the chunk.

After submission, the chunk is automatically vectorized and immediately becomes searchable.

Chunk operations

Each chunk supports the following actions:

Action	Description
Chunk Details	View full content and properties of the chunk.
View Original	Navigate to the original document context containing this chunk.
Delete	Immediately removes the chunk from the index—irreversible.
Enable/Disable	Toggle whether this chunk participates in search results.

Document operations

Each document in the document list supports the following operations:

Action	Description
View Chunks	Navigate to the chunk details page for this document.
Metadata	View or edit the document's metadata.
Tags	Add or manage tags for the document (accessible via the More menu).
Delete	Delete the document and all its associated chunks—irreversible (available in the More menu).
Batch Operations	Select multiple files using checkboxes at the top, then perform bulk actions.

API management

APIs related to document management:

• List documents
• Get file details
• Delete document