Knowledge base / RAG: document loading, template-based chunking, embedding indexing and hybrid (vector + keyword) retrieval with reranking. Depends on agent-start-model.

Workflow / agent orchestration: a DAG engine with a variable pool and real nodes (LLM, agent, knowledge retrieval, condition, HTTP, template, classifier, iteration). Depends on agent-start-model; knowledge retrieval is optional.

Enterprise agent runtime: pluggable strategies (ReAct, function-calling), persistent conversation memory, multi-agent delegation and human-in-the-loop tool approval. Depends on agent-start-model and agent-start-tools.

Ready-to-use REST web layer that exposes every module (model providers, knowledge, workflow, agent, tools, observability) over HTTP. Built on Spring MVC (servlet) for the request/response surface; token-level SSE for AI chat lives in ...

Spring Boot Starter —— WebFlux/Netty 侧的流式 AI 聊天层聚合。 引入本 starter 即等价于同时引入: - agent-start-completion OpenAI 兼容 /chat/completions/{appId} + Dify 兼容 /chat-messages (含 springdoc / swagger-webflux) ...

Runnable demo wiring all modules together; also hosts cross-module integration tests.

Standalone Spring Boot server: bundles every module and exposes /api/v1 REST + SSE endpoints. Run this to have a working backend without embedding the SDK into your own app.

Elasticsearch-backed VectorStoreFactory for agent-start-knowledge. One index per dataset.

Milvus-backed VectorStoreFactory for agent-start-knowledge. One collection per dataset.

PgVector-backed VectorStoreFactory for agent-start-knowledge. Requires PostgreSQL with the pgvector extension; activated by spring-agent.knowledge.vector-store=pgvector.

Spring Boot Starter —— MVC/Tomcat 侧的 REST CRUD 层聚合。 引入本 starter 即等价于同时引入: - agent-start-web REST CRUD 端点 (含 springdoc / swagger-webmvc) - agent-start-model 模型/凭证/预设 provider - agent-start-agent ...