spring-ai

Spring AI is an open-source application framework from the Spring ecosystem that brings Spring's design principles—portability, modularity, and strongly-typed APIs—to AI application development. Its core purpose is to bridge enterprise Data and APIs with AI Models through a consistent, Spring-friendly abstraction layer. Model provider support spans Anthropic, OpenAI, Amazon Bedrock, Google, Ollama, Mistral AI, DeepSeek and others, covering chat, embeddings, text-to-image, audio transcription, text-to-speech, and moderation. A portable API works across providers for both synchronous and streaming calls, while model-specific options remain accessible when needed. Vector store integration is equally broad, with support for Amazon Bedrock Knowledge Base, Amazon S3, Apache Cassandra, Azure Vector Search, Chroma, Couchbase, Elasticsearch, GemFire, MariaDB, Milvus, MongoDB Atlas, Neo4j, OpenSearch, Oracle, PostgreSQL/PGVector, Pinecone, Qdrant, Redis, Typesense, and Weaviate. A portable API across these stores includes a SQL-like metadata filter API. Key capabilities include Structured Outputs that map model responses to POJOs, Tool Calling so models can request client-side function execution for real-time information, an Observability module for AI operation insights, a document injection ETL framework for data engineering, and AI Model Evaluation utilities to assess generated content and guard against hallucinations. The ChatClient API provides a fluent, WebClient/RestClient-style interface for talking to chat models, while the Advisors API encapsulates recurring generative AI patterns and transforms data flowing to and from LLMs. First-class MCP (Model Context Protocol) support is available via Boot Starters and Java annotations, with STDIO, SSE, and Streamable-HTTP transports, letting applications consume MCP servers or expose Spring services to the AI ecosystem. Chat Conversation Memory is supported with pluggable persistent backends (JDBC, Cassandra, MongoDB, Neo4j, Redis), and Retrieval Augmented Generation (RAG) is built in. Spring Boot Auto Configuration and Starters cover all supported models and vector stores, and projects can be scaffolded through start.spring.io. Versioning aligns with Spring Boot: Spring AI 2.x targets Spring Boot 4.x, while Spring AI 1.1.x targets Spring Boot 3.5.x. The project ships reference documentation, Javadoc, upgrade notes, an examples repository, and a community organization for building Spring-based AI integrations. Building from source is optional and can be done with `./mvnw clean install` to publish artifacts to a local Maven repository.