Spring AI & RAG Course › Module 1 — Foundations › Spring AI Overview

Spring AI Overview

Module 1 · ~6 min read
What Spring AI is: a portable AI abstraction layer for Spring Boot. It wraps AI providers — Anthropic, Google Gemini, OpenAI, Ollama, and more — behind unified interfaces so your application code does not change when you switch providers.

The Problem Spring AI Solves

Every AI provider has its own SDK, its own request/response format, its own authentication scheme, and its own streaming protocol. Without an abstraction layer you end up tightly coupled to one provider, with vendor-specific code scattered throughout your application.

Spring AI provides a consistent programming model — the same ChatClient, EmbeddingModel, and VectorStore interfaces work regardless of which provider is underneath.

Key Abstractions

Abstraction Interface Purpose
ChatClient ChatClient Send prompts (user + system messages) and receive text or structured responses. Supports streaming.
EmbeddingModel EmbeddingModel Convert text strings into dense float vectors for semantic similarity comparison.
VectorStore VectorStore Store documents with their embeddings and search for semantically similar documents given a query.
DocumentReader DocumentReader Parse various file formats (PDF, Word, HTML) into Spring AI Document objects.
ImageModel ImageModel Generate images from text prompts (e.g., DALL-E, Imagen).

The "Write Once, Run Anywhere" Promise

Because all providers implement the same interfaces, you can swap Claude for Gemini by changing one Spring bean declaration. Your service code, your tests, your prompt logic — none of it changes.

In Power RAG, SpringAiConfig.java registers one ChatClient bean per provider. The service layer injects whichever bean it needs by @Qualifier. At runtime, the user can even choose a different provider per request — the service resolves the correct client dynamically. See Topic 05 and Topic 06 for the full pattern.

Your Application Code │ │ uses ▼ ┌──────────────────────────────────────────┐ │ Spring AI Abstractions │ │ ChatClient · EmbeddingModel · VectorStore│ └──────┬───────────┬──────────────┬────────┘ │ │ │ ▼ ▼ ▼ Anthropic Google Ollama Claude Gemini (local)

Spring AI BOM Pattern

Spring AI uses a Bill of Materials (BOM) to manage dependency versions. You import the BOM once in <dependencyManagement> and then declare individual starters without specifying versions — the BOM ensures all Spring AI artifacts stay in sync.

Power RAG uses Spring AI 1.1.2, the latest GA release for Spring Boot 3.x. See Topic 03: Project Setup for the full pom.xml configuration.

The BOM approach prevents version skew between Spring AI modules. Never declare individual Spring AI artifact versions — always let the BOM control them.

Provider Starters

Each provider has its own auto-configuration starter. Adding the starter to your pom.xml is all you need to get a working ChatModel bean wired up with your application.yml credentials:

← Previous Next →