AI Knowledge Base

AI Knowledge Base is an enterprise-level knowledge management system based on RAG (Retrieval-Augmented Generation) technology, supporting intelligent document retrieval and knowledge enhancement functions. It is responsible for document vectorization storage, semantic retrieval, content reranking, and real-time updates, providing precise knowledge retrieval and content management capabilities.

The AI Knowledge Base element hierarchical structure is Meta (rags.Meta) → Type (rags.NormalType) → Instance. Developers can quickly create AI Knowledge Base instance elements through JitAi's visual development tools.

Of course, developers can also create their own Type elements, or override the official rags.NormalType element provided by JitAi in their own App to implement their own encapsulation.

Supported Type Classes:

• rags.NormalType: Standard knowledge base type, providing complete document management and intelligent retrieval functions

Quick Start

Creating Instance Elements

Directory Structure

myKnowledgeBase/           # Knowledge base name (path customizable)
├── e.json                 # Element definition file
└── config.json            # Business configuration file

e.json File

e.json

{
  "title": "My Knowledge Base",
  "type": "rags.NormalType",
  "backendBundleEntry": "."
}

Business Configuration File

config.json

{
    "vectorModel": {
        "llmElement": "llms.LLMJitAppDevelop",
        "model": "text-embedding-v3"
    },
    "rerankModel": {
        "llmElement": "llms.LLMJitAppDevelop",
        "model": "gte-rerank-v2"
    },
    "topK": 10,
    "topN": 3,
    "vectorWeight": 0.5,
    "rerankType": "model",
    "vectorstoreType": "chroma"
}

Usage Example

Using AI Knowledge Base

# Get knowledge base instance
knowledge_base = app.getElement("rags.myKnowledgeBase")

# Query knowledge base
result = knowledge_base.query("What are the features of JitAi platform?")
print(result)

# Add document
knowledge_base.addDocumentByBusinessId(
    businessId="doc_001",
    document=[{"name": "document.pdf", "path": "/path/to/document.pdf"}]
)

Element Configuration

e.json Configuration

Parameter	Type	Native Type	Required	Description
title	Stext	str	Yes	Knowledge base display name
type	Stext	str	Yes	Fixed value: rags.NormalType
backendBundleEntry	Stext	str	Yes	Fixed value: .

Business Configuration File Configuration

Parameter	Type	Native Type	Required	Default	Description
vectorModel.llmElement	Stext	str	Yes	-	LLM element fullName
vectorModel.model	Stext	str	Yes	-	Vectorization model name
rerankModel.llmElement	Stext	str	No	-	Rerank LLM element fullName
rerankModel.model	Stext	str	No	-	Rerank model name
topK	Numeric	int	No	10	Number of results returned by vector retrieval
topN	Numeric	int	No	3	Final number of results returned after reranking
vectorWeight	Numeric	float	No	0.5	Vector retrieval weight, value between 0~1
rerankType	Stext	str	No	"model"	Rerank type: model/basic
vectorstoreType	Stext	str	No	"chroma"	Vector storage type

Methods

query

Intelligently query knowledge base content, retrieving relevant document segments based on semantic similarity.

Parameter Details

Parameter	Type	Native Type	Required	Description
query	Stext	str	Yes	User query content

Return Value

• Type: Ltext
• Description: Returns knowledge content related to the query, optimized through semantic retrieval and reranking

Usage Example

Query Knowledge Base

result = knowledge_base.query("What is JitAi platform?")
print(result)

addDocumentByBusinessId

Add new documents to the knowledge base, supporting custom chunking and cleaning configurations.

Parameter Details

Parameter	Type	Native Type	Required	Description
businessId	Stext	str	Yes	Business unique identifier for document management
document	File	list	Yes	Document attachment list, supports multiple formats
splitterConfig	JitDict	dict	No	Document chunking configuration
cleanerConfig	JitDict	dict	No	Content cleaning configuration

splitterConfig Configuration Items

Includes parameters such as chunk_size (chunk size), chunk_overlap (overlap length), separator (delimiter), etc.

cleanerConfig Configuration Items

Includes options such as remove_extra_whitespace (remove extra whitespace), normalize_unicode (Unicode normalization), remove_urls (remove URLs), etc.

Return Value

• Type: JitDict
• Description: Contains document processing result information

Usage Example

Add Document to Knowledge Base

result = knowledge_base.addDocumentByBusinessId(
    businessId="product_manual_v1.0",
    document=[{"name": "manual.pdf", "path": "/path/to/manual.pdf"}],
    splitterConfig={"chunk_size": 500, "chunk_overlap": 100},
    cleanerConfig={"remove_extra_whitespace": True, "normalize_unicode": True}
)

deleteDocumentByBusinessId

Delete specified documents and their related vector data from the knowledge base.

Parameter Details

Parameter	Type	Native Type	Required	Description
businessId	Stext	str	Yes	Business identifier of the document to be deleted

Return Value

• Type: boolean
• Description: Returns true if deletion is successful, false if failed

Usage Example

Delete Document

success = knowledge_base.deleteDocumentByBusinessId("product_manual_v1.0")
print(f"Deletion result: {success}")

queryKeywords

Query the knowledge base based on keyword lists, supporting exact matching and relevance ranking.

Parameter Details

Parameter	Type	Required	Description
keywords	list	Yes	Keyword list, supports up to 5 keywords
limit	int	No	Limit return count, defaults to configured topN value

Return Value

• Type: str
• Description: Matched document content, connected by line breaks, including metadata information such as page tags, document titles, document types, etc.

Usage Example

Keyword Query

# Single keyword query
result = knowledge_base.queryKeywords(["refund"])
print(result)

# Multi-keyword query, limit to 3 results
result = knowledge_base.queryKeywords(
    keywords=["refund", "application", "process"], 
    limit=3
)
print(result)

Properties

None

Advanced Features

Intelligent Semantic Retrieval

Implements semantic understanding based on vector embedding technology, supporting intelligent retrieval with non-exact matching.

Document Chunking Processing

Supports flexible document chunking strategies, optimizing retrieval accuracy through parameters such as chunk_size and chunk_overlap.

Content Cleaning Optimization

Provides multiple content cleaning options, including whitespace character processing, Unicode normalization, etc., to improve document quality.

Reranking Algorithm

Adopts a two-stage retrieval strategy, first vector recall of candidate results, then precise ranking through reranking models.

Real-time Content Updates

Supports incremental updates and precise deletion based on businessId, ensuring real-time content and version management of the knowledge base.