Create a Knowledge Base

A Knowledge Base is the core entity in the RAG platform. It groups documents, Chunks, embeddings, and index configuration together and exposes retrieval and Q&A capabilities through the API.

Knowledge Base types

When you create a Knowledge Base, choose a type. Each type uses a different retrieval method:

Type	Data	Retrieval method	Notes
Unstructured Documents	PDF, Word, Markdown, HTML	Hybrid vector + keyword	For text-heavy documents.
Structured data	CSV, Excel, RDS tables	Natural-language-to-SQL	For precise numerical queries.
Image library	Product photos, design files	Image-to-image search, visual Q&A	For image search.
Audio / video library	Screen recordings, training videos	Transcription + segment locating	Auto-transcribes and indexes by segment.

Create from the console

Open Index Management

Sign in to the console, click Index Management in the left navigation, then click New Knowledge Base in the top right.

Fill in basic info

Field	Required	Notes
Name	Yes	1–50 characters, unique within the workspace.
Description	No	Up to 200 characters, used in filtering.
Type	Yes	Unstructured / Structured / Image / Audio-video.

Pick a parsing method

Parsing method	Description	Best for
Native document parsing	Standard text extraction.	Clean electronic documents.
Document Intelligence Parsing	Layout-aware OCR + structure recovery.	Scans and complex layouts.
LLM-based parsing	An LLM interprets the document structure.	Non-standard formats.
Qwen-VL parsing	Visual language model.	Documents that mix text and images.
Audio / video parsing	Speech-to-text with timestamp locating.	Audio and video files.

Configure the Chunking Strategy

Parameter	Default	Range	Notes
Chunking method	Smart Chunking	Smart / Fixed-length / Heading-based	Smart Chunking splits along the document layout.
Chunk size	1024	256–4096	Maximum length of a single Chunk.
Chunk overlap	10%	0%–50%	Overlap between adjacent Chunks.

See Chunking and Embedding for details.

Create

Click Confirm. The Knowledge Base enters the Ready state and you can start uploading documents.

Create through the API

Use the Create Knowledge Base and Import endpoint to create a Knowledge Base and submit an import job in one call:

curl -X POST https://{workspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/indices/rag/index/create_v2 \
  -H "Authorization: Bearer $DASHSCOPE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "index_name": "Product Manual",
    "sink_type": "DEFAULT",
    "structure_type": "unstructured",
    "embedding_type": "embedding_v3",
    "source_type": "DATA_CENTER_FILE",
    "document_ids": ["file-xxx-001", "file-xxx-002"]
  }'

Limits

Item	Limit	Notes
Knowledge Bases per workspace	50	Submit a ticket to request a higher limit.
Documents per Knowledge Base	100,000	Split larger collections.
Per-document size	100 MB	Split larger files.

You can change the Chunking Strategy after creation, but changing the embedding model triggers a full index rebuild.

Once the Knowledge Base is created, move on to Upload Documents.

​Knowledge Base types

​Create from the console

​Create through the API

​Limits

Knowledge Base types

Create from the console

Create through the API

Limits