spring-projects/spring-ai
Spring Ai
Spring AI is an application framework for Java that provides a portable, fluent API for integrating AI models, tools, and vector stores into applications. It wraps multiple AI providers behind a common interface, allowing developers to switch between chat, embedding, image, and speech models without changing application code. The framework includes a chainable chat client API similar to WebClient or RestClient, supports both synchronous and streaming interactions, and offers structured output conversion that transforms unstructured AI responses into strongly-typed Java objects.
The framework distinguishes itself through its tool calling capabilities, which allow AI models to invoke client-side Java methods during conversations, with support for dynamic resolution, lifecycle management, and annotation-driven registration. It implements the Model Context Protocol (MCP) to establish a standardized bridge between AI models and external services for tool and resource discovery. Spring AI also provides a retrieval-augmented generation (RAG) pipeline that combines document retrieval from vector stores with language model generation, and an advisor framework that encapsulates recurring patterns like memory, tool-calling, and RAG into reusable interceptors.
Beyond core AI interactions, the framework handles document processing by loading, chunking, and enriching documents from structured files, and manages vector embeddings across multiple database providers with metadata filtering. It includes content moderation capabilities to inspect AI output against safety policies, and provides observability through logging, tracing, and evaluation utilities that assess response quality and detect hallucinations. The framework supports audio transcription, image generation, and speech synthesis through portable APIs, and manages conversation history across sessions using pluggable persistent backends.
Spring AI bootstraps AI model and vector store connections automatically using Spring Boot starters and dependency-driven configuration, with project skeletons available through Spring Initializr.
Features
- AI Application Frameworks - Ships a portable, fluent Java framework for integrating AI models, tools, and vector stores into applications.
- Fluent Chat Clients - Provides a chainable, idiomatic API for constructing and sending chat requests to AI models.
- Tool Call Execution Loops - Handles the iterative cycle of invoking tools requested by the model and feeding results back until no more calls are needed.
- Provider Switching - Provides a portable abstraction to swap between AI providers like OpenAI and Ollama without code changes.
- Model Context Protocol - Implements the Model Context Protocol to establish a standardized bridge between AI models and external services.
- Model Context Protocol Integrations - Establishes a standardized bridge between AI models and external services using the Model Context Protocol.
- Chat Client Advisors - Adds a chain of interceptors to modify prompts, inject context, or log interactions before calling the AI model.
- Fluent Chat Client APIs - Ships a fluent, chainable API for constructing and sending chat requests to AI models with configurable advisors.
- Multi-Provider Chat Clients - Connects Java applications to multiple AI chat providers through a unified, portable API with streaming and structured output support.
- Portable Chat APIs - Provides a portable, fluent API to send prompts and receive responses from major AI providers in both sync and streaming modes.
- Streaming Chat Responses - Streams the model's reply token-by-token using a reactive Flux API for real-time output.
- AI Model Clients - Provides a fluent, chainable API for constructing and sending chat requests to AI models.
- Prompt Template Retrieval - Constructs user and system messages using template strings with runtime variable substitution.
- AI Tooling Frameworks - Registers Java methods as callable tools that AI models can invoke during conversations with dynamic resolution.
- AI Tooling Interfaces - Registers Java methods or functions as callable tools that an AI model can invoke during a conversation.
- AI Tool Annotations - Exposes Java methods as callable AI tools by annotating them with @Tool, with automatic input schema generation.
- Conversation Memory Stores - Stores and retrieves chat history for coherent multi-turn conversations across sessions.
- Retrieval-Augmented Generation - Combines document retrieval with language model generation for context-aware question answering.
- Model Output Transformers - Transforms unstructured AI responses into strongly-typed Java POJOs for structured data handling.
- MCP Protocol Integrations - Establishes a standardized bridge between AI models and external services using the Model Context Protocol.
- Function Calling Interfaces - Provides a framework for AI models to request execution of user-defined functions to access real-time information.
- Automatic Tool Executions - Uses an advisor to handle the full tool-calling lifecycle automatically without manual intervention.
- Advisor Frameworks - Wraps recurring patterns like memory, tool-calling, and RAG into reusable advisors that transform data sent to and from language models.
- Reusable Advisor Patterns - Wraps recurring patterns like memory, tool-calling, and RAG into reusable advisors that transform data flow.
- Tool Calling - Exposes Java methods as callable tools that an AI model can invoke during a conversation.
- Persistent Chat Histories - Stores and retrieves conversation history across sessions using pluggable persistent backends.
- Service Provider Abstractions - Wraps multiple AI model providers behind a common interface, allowing provider switching without code changes.
- Multi-Provider Abstractions - Provides a consistent abstraction for interacting with chat, embedding, image, and speech models across multiple providers.
- Model Response Streaming - Supports both synchronous and streaming API options for AI model interactions with real-time output.
- Textual Response Relevance Evaluators - Provides built-in evaluators that judge whether AI responses stay on topic and are factually supported.
- Retrieval-Augmented Generation Frameworks - Implements RAG pipelines that combine document retrieval from vector stores with language model generation.
- Retrieval Augmented Generation Pipelines - Builds question-answering systems that retrieve relevant documents from a vector store and inject them into AI prompts.
- Prompt Augmenters - Enriches user queries with relevant information from a knowledge base before sending them to the AI model.
- Structured Output Converters - Transforms unstructured text responses from AI models into strongly-typed Java objects using built-in converters.
- Text Embeddings - Converts text, documents, or batches of text into numerical vector representations using an embedding model.
- AI Model and API Integration - Provides a consistent abstraction for interacting with chat, text-to-image, and embedding models across multiple providers.
- AI Interaction Interceptors - Intercepts, inspects, and modifies prompt data and responses flowing between the user and the language model through advisors.
- Knowledge Base Question Answering - Implements a Retrieval-Augmented Generation pattern that uses a vector store to retrieve relevant documents and answer user questions.
- Structured Return Objects - Maps AI model text output directly into Java objects or collections of objects for type-safe use.
- Document Ingestion Pipelines - Extracts, transforms, and loads documents into a vector store for retrieval-augmented generation pipelines.
- Vector Stores - Offers a portable API to store and query vector embeddings across a wide range of database providers.
- Conversation Memory Retrieval - Supports chat conversation memory and retrieval-augmented generation for context-aware interactions.
- Factual Grounding - Checks if claims in AI responses are factually supported by the given context to catch hallucinations.
- Conversation History Backends - Stores and retrieves conversation history across multiple pluggable storage backends.
- Embedding Generation - Provides a portable API for converting text into numerical vector representations across multiple AI providers.
- MCP Server Integrations - Supports the Model Context Protocol for consuming external AI services or exposing Spring services to the AI ecosystem.
- AI Prompt Sending APIs - Sends a textual prompt to an AI language model and receives a generated natural-language response.
- AI Advisor Chains - Implements a configurable chain of advisor interceptors that modify prompts and responses around AI model calls.
- Object Mappings - Converts structured AI model responses directly into Plain Old Java Objects for type-safe downstream processing.
- POJO Mappings - Converts AI model responses directly into Plain Old Java Objects for type-safe downstream processing.
- Chat API Integrations - Offers a fluent, idiomatic API for interacting with chat models, similar to WebClient and RestClient.
- Model Tool Calls - Allows AI models to request execution of user-defined tools during conversations for real-time data access.
- Anthropic Claude Connections - Connects to Anthropic's Claude models through the official Java SDK for chat and completion tasks.
- Dynamic Tool Management - Provides a ToolCallbackResolver for looking up tools by name at runtime from a registry.
- Client Connection Lifecycles - Handles protocol negotiation, tool discovery, resource access, and message transport for MCP server connections.
- Chat Model Configurations - Sets model parameters such as temperature, token limit, and stop sequences at startup or per request.
- Service Exposures - Makes Spring-based services discoverable and callable by AI applications through the Model Context Protocol.
- Text-to-Image Generators - Provides a unified API for generating images from text descriptions using supported AI models.
- API Speech Synthesizers - Provides a portable API for generating spoken audio from text using supported AI providers.
- Bean Deserializations - Converts raw AI model text responses into typed Java objects by appending JSON schema instructions to the prompt.
- Native Schema Enforcements - Sends JSON schemas directly to the model's structured output endpoint for reliable schema compliance.
- Full Response Aggregation - Retrieves the complete AI model response including metadata such as token usage and multiple generations.
- Image Model Provider Switching - Provides a portable API to switch image generation providers by changing a configuration value.
- Transcription Provider Switching - Provides a portable API to switch transcription providers by changing a configuration value.
- Image Description Generators - Provides a unified API for generating images from text prompts using supported AI models.
- Unified Audio Transcription - Provides a portable API for converting spoken audio into text using supported AI providers.
- Streaming Speech Outputs - Provides streaming speech synthesis that delivers audio incrementally for real-time playback.
- Vector Database ETL Tools - Extracts, transforms, chunks, and loads documents from various sources into vector databases for RAG.
- Similarity Query Engines - Accepts text queries and returns semantically similar documents ranked by vector similarity.
- AI Observability and Evaluation - Captures metrics, traces, and logs for AI model interactions and evaluates response quality to detect hallucinations.
- Document Chunking Expansion - Divides document text into smaller segments based on token count or structure for AI context windows.
- Document and File Processing - Reads PDF, JSON, HTML, Markdown, DOCX, PPTX, and plain text files into Document objects.
- Document Data Extraction - Loads, processes, and structures documents from various sources for ingestion into AI workflows.
- Data Loaders - Extracts, transforms, and loads data from various sources into a vector database to support retrieval-augmented generation.
- Document Writers - Stores processed Document objects into a vector database for use in retrieval-augmented generation workflows.
- Filtered Queries - Offers a portable API across vector database providers, including a SQL-like metadata filter for precise retrieval.
- Metadata Filters - Applies portable SQL-like metadata filters to vector store queries for precise result retrieval.
- Filtered Similarity Searches - Restricts vector similarity results using metadata field expressions for precise retrieval.
- Tool and Resource Exposures - Provides tools, resources, and prompt templates to MCP clients with concurrent connection management.
- TTS Provider Switching - Provides a portable API to switch text-to-speech providers by changing a configuration value.
- MCP Tool Behavior Annotations - Creates tools, resources, and prompts on a server using Java annotations to reduce boilerplate.
- Prompt Templates - Builds user and system messages by substituting runtime variables into template strings at request time.
- Audio-to-Text Alignment - Provides a portable API for converting spoken audio from files into written text using speech-to-text models.
- Text-to-Speech Synthesizers - Provides a portable API for converting text strings into audio files using text-to-speech models.
- Content Moderation - Inspects AI model output against safety policies to block harmful or inappropriate responses.
- Auto-Configuration Mechanisms - Bootstraps AI model and vector store connections automatically using Spring Boot auto-configuration.
- AI Starter Configurations - Auto-configures AI model and vector store connections using dependency-driven starters and standard properties.
- Spring Boot AI Starters - Auto-configures AI model and vector store connections using Spring Boot starters and dependency-driven configuration.
- AI Observability - Provides insights and monitoring data into the lifecycle of AI model interactions for debugging and performance analysis.
