RAG Knowledge Base

io.agentscope.core.rag.Knowledge is AgentScope’s interface for plugging in an external knowledge base. The Agent uses it during inference to retrieve documents that are then handed to the model. The agentscope-extensions-* repository ships several implementations:

Extension	Type	Best for
Simple	Self-managed: embeddings + vector store	Bring-your-own vector store (PgVector / Milvus / Qdrant / Elasticsearch / in-memory)
Bailian	Alibaba Cloud Bailian Knowledge Base	Use a Bailian-hosted enterprise KB
Dify	Dify dataset	Already maintaining KB content in Dify
HayStack	Self-hosted HayStack RAG	Existing HayStack pipelines
RAGFlow	RAGFlow service	Complex documents needing OCR / knowledge graphs

Same wiring everywhere

ReActAgent agent = ReActAgent.builder()
    .name("Assistant")
    .model(model)
    .knowledge(knowledge)        // any Knowledge implementation
    .ragMode(RAGMode.AGENTIC)    // or STATIC, NONE
    .build();

You can also expose retrieval as a tool that the Agent calls explicitly:

KnowledgeRetrievalTools tools = new KnowledgeRetrievalTools(knowledge);
Toolkit toolkit = new Toolkit();
toolkit.registerObject(tools);

Choosing one

Need	Recommendation
Full control over embeddings + store	Simple
Alibaba Cloud ecosystem, enterprise hosting	Bailian
Team already curates content in Dify	Dify
Complex ETL (PDF tables, OCR, knowledge graphs)	RAGFlow
Existing HayStack RAG pipeline	HayStack

Aside from Simple, the platform-backed implementations are retrieval only: document ingestion / updates happen via the platform console or its native API. This keeps Knowledge implementations swappable on the Agent side.