Simple Knowledge¶

agentscope-extensions-rag-simple is the “DIY end-to-end” RAG implementation: it bundles document readers, chunking strategies, embedding adapters, and five out-of-the-box vector store adapters.

Use it when: you’re happy to run embeddings + vector store yourself and don’t want a third-party RAG platform.

Add the dependency¶

<dependency>
    <groupId>io.agentscope</groupId>
    <artifactId>agentscope-extensions-rag-simple</artifactId>
    <version>${agentscope.version}</version>
</dependency>

Quickstart¶

import io.agentscope.core.embedding.dashscope.DashScopeTextEmbedding;
import io.agentscope.core.rag.knowledge.SimpleKnowledge;
import io.agentscope.core.rag.store.InMemoryStore;
import io.agentscope.core.rag.model.RetrieveConfig;

// 1) Embedding model
EmbeddingModel embeddings = DashScopeTextEmbedding.builder()
    .apiKey(System.getenv("DASHSCOPE_API_KEY"))
    .modelName("text-embedding-v3")
    .dimensions(1024)
    .build();

// 2) Vector store (in-process here)
VDBStoreBase store = InMemoryStore.builder().dimensions(1024).build();

// 3) Assemble Knowledge
SimpleKnowledge knowledge = SimpleKnowledge.builder()
    .embeddingModel(embeddings)
    .embeddingStore(store)
    .build();

// 4) Ingest documents
List<Document> docs = new TikaReader().read(input).block();
knowledge.addDocuments(docs).block();

// 5) Retrieve
List<Document> hits = knowledge.retrieve(
    "What is AgentScope?",
    RetrieveConfig.builder().limit(5).scoreThreshold(0.5).build()
).block();

Built-in document readers¶

The io.agentscope.core.rag.reader package contains readers for common formats; each produces List<Document>:

Reader	Input
`TextReader`	Plain text
`PDFReader`	PDF (PDFBox-backed)
`WordReader`	Microsoft Word documents
`ImageReader`	Images, paired with multimodal embeddings
`TikaReader`	Generic Apache Tika fallback
`ExternalApiReader`	External parser APIs (OCR / custom pipelines)

The resulting Document objects already carry metadata; pair with TextChunker and SplitStrategy for chunking.

Built-in embedding providers¶

Class	Service	Mode
`DashScopeTextEmbedding`	Alibaba Cloud DashScope	Text
`DashScopeMultiModalEmbedding`	Alibaba Cloud DashScope	Multimodal (text/image)
`OpenAITextEmbedding`	OpenAI-compatible API	Text
`OllamaTextEmbedding`	Local Ollama	Text

Implement EmbeddingModel to add your own.

Built-in vector stores¶

Implementation	Deployment
`InMemoryStore`	In-process (dev / testing)
`PgVectorStore`	PostgreSQL + pgvector
`MilvusStore`	Milvus
`QdrantStore`	Qdrant
`ElasticsearchStore`	Elasticsearch (`dense_vector`)

Switching stores is a one-line change: pass a different VDBStoreBase to SimpleKnowledge.builder().embeddingStore(...).

Retrieval parameters¶

RetrieveConfig controls retrieval:

Field	Notes
`limit`	Top-K
`scoreThreshold`	Minimum score (0–1)
`metadata`	Filter by document metadata

Wire into an Agent¶

ReActAgent agent = ReActAgent.builder()
    .name("Assistant")
    .model(model)
    .knowledge(knowledge)
    .ragMode(RAGMode.AGENTIC)   // Agent decides when to retrieve
    .build();