acorn-springai

Acorn with Spring AI

A Spring AI module that integrates Acorn with Spring AI for tooling and chat memory. It provides:

• Converts GraphQL API into callback functions for LLM tooling
• Implements ChatMemory against GraphQL API

Example Spring AI Application using Acorn

Ricky and Morty Chatbot

TODO

Credit Card Agent

The implementation expects that the DataSQRL rewards credit card example is running for information retrieval and message persistence.

 docker run -it -p 8888:8888 -p 8081:8081 -p 9092:9092 --rm -v $PWD:/build datasqrl/cmd:dev run -c package-rewards-local.json

You can then run the Spring boot application with the following settings:

• You need to set the environment variable SPRING_AI_OPENAI_API_KEY to your OPENAI API key
• You need to set backedUrl config option Spring to connect to the running DataSQRL project. On the command line you can do --config.backendUrl=http://localhost:8888/graphql.

Open your browser and ask messages directly via the REST endpoint:

http://localhost:8080/agent/4?message=what-did-you-last-tell-me

• The number 4 is for the customerid (1-9)
• The message is the customer message

TODOs

• Need to update the AcornChatMemory to save and retrieve toolCalls. Those need to be mapped onto InputTypes in the GraphQL API to support their structure (see record ToolCall in AssistantMessage - it has 4 string fields). So far, I have not been able to observe a message with tool calls in the advisor, even though a tool was called. This needs to be troubleshot first.
• Need to move out the wrapper classes around acorn-graphql into a separate acorn-springai module so this can be reused more widely. Those are currently in the acorn package.
  • Need to add a helper configuration builder class to create the tools and AcornChatMemory. Allow instantiating from string and Resource. Check the current Config implementation for details. We want this to be simpler.
    • First, create APIExecutor via builder on SpringGraphQLExecutor (or multiple as needed)
    • Second, create AcornChatMemory via builder that hides APIChatPersistence
    • Third, create tools for each GraphQL schema. Create builder for easier use.