Skip to main content

Document Management and Parsing

Upload documents, inspect parsing results, and manage Chunks.

Documents are the content source for a Knowledge Base. Once uploaded, the platform parses them, splits them into Chunks, embeds them, and builds the Index automatically.

Upload documents

1

Open the Knowledge Base

In Index Management, click the name of the target Knowledge Base.
2

Upload files

Click Upload Documents and select local files. Up to 50 files per call; each file must be under 150 MB.
3

Wait for processing

Document status transitions through:Uploading → Parsing → Chunking → Embedding → ReadyOnce status is Ready, the document is searchable.

Supported formats

FormatExtensionsNotes
PDF.pdfNative PDFs are extracted directly; scans go through auto OCR.
Word.docx, .docPreserves heading and paragraph hierarchy.
PowerPoint.pptxSplits by slide.
Excel / CSV.xlsx, .xls, .csvSplits by row; first row provides field names.
Markdown.mdSplits along heading levels.
HTML.htmlStrips scripts and styles automatically.
Plain text.txtUTF-8 encoded.

Document status

StatusMeaningAction
UploadingFile transfer in progressWait.
ParsingExtracting document structure and textWait.
ChunkingSplitting into ChunksWait.
EmbeddingGenerating embedding vectorsWait.
ReadySearchableNone.
FailedParsing or embedding errorCheck the error details, then re-upload after fixing.

Parsing methods

The platform offers several parsing methods. Pick one when creating the Knowledge Base; you can also change it in Knowledge Base settings later.
MethodHow it worksStrengthsBest for
Native document parsingStandard text extractionFast, low costClean electronic documents.
Document Intelligence ParsingLayout-aware OCR + table recognition + formula recoveryHigh-fidelity layout reproductionScans, reports, contracts.
LLM-based parsingAn LLM interprets the document's semantic structureHandles non-standard formatsMixed-layout documents.
Qwen-VL parsingVisual language modelUnderstands text-image contentTechnical documents with diagrams.
For scans or complex layouts, use Document Intelligence Parsing. For plain-text formats like Markdown and TXT, native document parsing is enough.

Chunk management

From the Knowledge Base detail page you can view, edit, and manage every Chunk. Each Chunk shows:
  • Chunk content and boundaries.
  • Chunk length.
  • Embedding status.
  • Source document (use the All files filter to browse Chunks across the entire Knowledge Base in one view).

Create a Chunk

Click Create Chunk and fill in:
FieldConstraintDescription
Source documentRequiredPick the document this Chunk belongs to.
Chunk title0–50 charactersShown in the Chunk list.
Imagepng / jpg / bmp / gif, ≤ 20 MB per imageOptional, for text-with-image Chunks.
Chunk content0–6000 charactersChunk body text.
After submission the Chunk enters the embedding queue and becomes searchable once embedding finishes.

Edit or delete a Chunk

Every Chunk can be edited or deleted individually. Edits trigger automatic re-embedding; deletion removes the Chunk from the index immediately and is irreversible.
The Chunk entry point varies slightly across workspaces — some show the Chunk table directly on the detail page, others surface it through a dedicated Chunks entry. Follow the actual layout of your console.

Management actions

ActionDescription
View file detailsClick a file name to see parsing results and the Chunk list.
Delete documentRemoves the document along with all of its Chunks and vectors — irreversible.
Re-parseAfter changing the parsing method, re-run parsing on existing documents.
Bulk actionsSelect multiple files to delete or re-parse them in bulk.
Document deletion is irreversible. Previously returned retrieval results are unaffected, but later queries no longer match Chunks from the deleted document.

API management

Document management APIs:
After uploading documents, tune your chunking and embedding strategy to improve retrieval quality.
Getting Started
Playground
Data Ingestion
Knowledge Service
Cookbook
Integration
Settings, Billing & Limits