Model management: provider/credential registry, model instance factory and a Spring AI based invocation layer (chat, embedding, rerank). Importable standalone.

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.

Tool / connector ecosystem: a Tool SPI, built-in tools, a Spring AI ToolCallback adapter and an MCP client provider — the n8n-style integration layer agents and workflows call into.

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.

Observability / LLMOps: per-call token + cost + latency metering for every LLM invocation, persisted for aggregation, with optional Micrometer metrics. Plugs into the model module via ChatModelDecorator.

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) ...

DeepSeek model provider using the official spring-ai-deepseek SDK. Adding this jar transparently upgrades the built-in "deepseek" preset from the generic OpenAI-compatible client to the native DeepSeek client, which surfaces ...

Volcengine Ark (Doubao) model provider. Uses the OpenAI-compatible REST protocol under the hood but exposes an explicit "endpointId" credential field — Volcengine identifies inference endpoints by ep-xxx ids rather than model ...

Zhipu AI (GLM) model provider using the official spring-ai-zhipuai SDK. Adding this jar transparently upgrades the built-in "zhipu" preset from the generic OpenAI-compatible client to the native Zhipu client (which supports ...