Unlock Spring AI: Build Java Generative Apps with Model Switching, Memory, and Prompt Engineering

What is Spring AI?

Spring AI is a Java library that simplifies the development of applications with generative AI features, much like Spring Boot simplifies Spring applications. It aims to reduce the complexity of integrating AI models, data sources, and APIs, making Java a first‑class language for AI development.

Supported Model Providers and Types

Spring AI supports the major large‑model providers, including Anthropic, OpenAI, Microsoft, Amazon, Google, and Ollama. The library abstracts the following model capabilities:

Chat Completion

Embedding

Text‑to‑Image

Audio Transcription

Text‑to‑Speech

Content Moderation

These providers can be switched at runtime with second‑level latency, greatly improving development efficiency.

Getting Started: Hello‑World Chatbot

The example uses Spring AI 1.0 with the openai and ollama starters to demonstrate dynamic model switching. It is important to avoid the 1.1.0‑M7 snapshot because its Maven dependencies cannot be resolved.

A configuration class imports the required starters and defines the model selection. The controller class contains a method that calls the model and returns the response. Running the application produces a simple “Hello World” response from the chosen AI model.

Conversational Memory

Conversational Memory enables the system to retain context across multiple turns, allowing coherent multi‑turn interactions. The demo adds a conversationId parameter to link requests to the same memory store.

When the same conversation ID (e.g., 1) is used, the model correctly remembers the previous question. Changing the ID to a new value (e.g., 2) resets the memory, demonstrating the dependency on the conversation identifier.

Beyond in‑memory storage, Spring AI also offers MySQL, PostgreSQL, Cassandra, Neo4j, and other persistence options for memory data.

Prompt Engineering

Prompt Engineering bridges human intent and model output by crafting precise instructions. It involves four elements—role, task, context, and output format—and aims to clarify intent, control output format, reduce hallucinations, and adapt to specific domains such as healthcare or finance.

The demo implements a completionPrompt method with three steps:

Create a prompt template that includes role, task, context, and desired output.

Inject variable parameters into the template.

Invoke the model and return the result.

Sample runs show that providing the parameter “洞见” yields a modern‑style Chinese paragraph, while the parameter “李白” produces a classical‑style Chinese text, illustrating how prompt design influences output style.

Conclusion

The article covers the basics of Spring AI, including model support, conversational memory, and prompt engineering. Future posts will explore Retrieval‑Augmented Generation (RAG), multimodal capabilities, Model‑Centric Programming (MCP), and vector‑store integrations.