Spring AI in action

This project showcases the use of Spring AI to bring AI features to your Spring application.

Using Spring AI, you can leverage a wide range of AI providers (such as OpenAI, Mistral AI, Google Gemini and many others) along with additional services such as vector databases.

Spring AI provides an abstraction layer atop AI providers API: this way you can focus on your code by using the same API, no matter which AI provider you pick.

Spring AI is to AI providers what Spring Data is to databases: the best abstraction layer to build portable Spring apps.

This repository includes a few demos you can run to learn about Spring AI.

How to build and run this app?

You need Java 21+ to build and run this app.

Use this command to build the app:

./mvnw clean package

Before running this app, make sure you have Docker daemon running, along with Docker Compose.

Use this command to bootstrap a Redis database (which is used by some demos available in this app):

docker compose up

You can run this app using this command:

./mvnw spring-boot:run

The app will automatically bind to the Redis database running as a container.

AI providers

This application has been tested with 3 AI providers:

• OpenAI with gpt4o
• Azure OpenAI with gpt4o
• Mistral AI with open-mixtral-8x22b
• Ollama (local) running Mistral AI with model mistral

Please note that those models are not created equal: some features (such as function calling) may not properly work depending on the model you pick.

You may run this app with additional AI providers and/or models, providing you add the right Spring AI dependency provider in the file pom.xml along with the configuration.

OpenAI

You need an OpenAI API key to run this app. Refer to this page to create and get an API key.

Set your API key as an environment variable:

export OPENAI_API_KEY=xxxxxx

Set openai as the AI provider:

export APP_AI_PROVIDER=openai

Azure OpenAI

You need an Azure OpenAI API key to run this app. The deployment name should be gpt-4o Set your API key as an environment variable:

export AZURE_OPENAI_API_KEY=xxxxxx

Set azure as the AI provider:

export APP_AI_PROVIDER=azure

Mistral AI

You need a Mistral AI API key to run this app. Refer to this page to create and get an API key.

Set your API key as an environment variable:

export MISTRALAI_API_KEY=xxxxxx

Set mistralai as the AI provider:

export APP_AI_PROVIDER=mistralai

Ollama (local)

Using Ollama you can run a LLM on your workstation. Follow these instructions to download and install Ollama.

Start Ollama using this command:

ollama serve

Use this command to download the Mistral AI model for Ollama

ollama pull mistral

OpenWeatherMap

This app relies on OpenWeatherMap to get live weather information for a given city.

Prior to running this app, you need to set the OPENWEATHERMAP_API_KEY environment variable.

See this page to create an OpenWeatherMap account for free and get an API key.

Use this command to set the environment variable:

export OPENWEATHERMAP_API_KEY=xxxxx

🎉 You're ready to go!

Demos

Chat API

Basic use of the Spring AI ChatClient:

http ":8080/chat/v1?q=How are you today?"

Prompt templating:

http ":8080/chat/v2?topic=devops"

Stateless chat conversations:

http ":8080/hello"

Stateful chat conversations (keep track of previous messages):

http ":8080/hello-memory"

Converting chat responses to Java entities:

http ":8080/chat/v3?topic=devops"

Same as above with content negotiation (prioritizing XML output):

http ":8080/chat/v3?topic=devops" Accept:text/xml

Transform a text into a set of Nodes & Relationships:

http ":8080/chat/v4"

Retrieval Augmented Generation (RAG)

Load and index data (this task actually processes input files in background):

http ":8080/movies/init"

Create a new movie using 2 movie titles:

http ":8080/movies?titles=Lord of the Rings,Despicable Me"

Same as above but using genre drama (default is comedy):

http ":8080/movies?titles=Furiosa,Snow White&genre=drama"

Same as above but translating the result into french:

http ":8080/movies?titles=Ghostbusters,Despicable Me&genre=drama&lang=fr"

Clear the database (make sure you restart the app after running this command):

redis-cli flushdb

Image Generation

For this section please enable OpenAI as the code uses DALL-E image model. You may use other AI providers as long as you pick the right image model.

Generate an image:

open "http://localhost:8080/imagegen?q=A vibrant photograph of a Java programmer working in his office with the Eiffel tower in background, wide shot, soft focus" 

Function Calling

Verify that the AI provider has no way to figure out current weather if no function is registered:

http ":8080/chat/v1?q=Should I bring my sunglasses when I'll get to Paris later today?"

Call an inline function:

http ":8080/weather/v1?city=Paris"

Call a function and map the result to a Java entity:

http ":8080/weather/v2?city=Paris"

Using functions, verify that the AI provider can now answer weather related questions:

http ":8080/weather/v3?q=Should I bring my sunglasses when I'll get to Paris later today?"

Same as above but using a complex query involving many cities:

http ":8080/weather/v3?q=Sort the top 5 most populated cities in France by temperature."

Multimodal Prompts

Combining text, image and functions to answer a question, mapping the result to a JSON structure:

http ":8080/weather/v4?u=https://live.staticflickr.com/65535/54135349734_8503393ea0_h.jpg"

Contributing

Contributions are always welcome!

Feel free to open issues & send PR.

License

Copyright © 2024 Broadcom, Inc. or its affiliates.

This project is licensed under the Apache Software License version 2.0.