Add TOON (Token-Oriented Object Notation) Support

[maksymuimanov]

trafficstars

Expected Behavior

Spring AI should support the TOON format as a serialization option when sending structured data to LLM models.

Example usage:

ExampleObject exampleObject = new ExampleObject();

String response = chatClient.prompt()
    .mapper(toonMapper) // <-- the TOON mapper bean
    .user(exampleObject)
    .call()
    .content();

This would allow applications to use TOON instead of JSON/YAML for model inputs, improving efficiency and reducing token count.

Current Behavior

Spring AI currently supports JSON and YAML (via built-in or user-provided serializers), but there’s no native TOON support. Developers must manually serialize/deserialize TOON data and integrate it into prompts or function-calling logic.

Context

• TOON (Token-Oriented Object Notation) is a compact, token-efficient format designed for LLM input/output.
• Article reference:
  • https://codefarm0.medium.com/toon-a-token-friendly-data-format-for-the-llm-era-with-java-code-examples-4dcafbfb03ce
  • https://blog.arun.im/toon4j-slash-your-llm-token-costs-by-50-with-this-java-library-1e16578ff9eb
• Benefits of adding TOON support:
  • Reduces token usage and improves prompt window efficiency.
  • Provides native serialization/deserialization similar to JSON/YAML.
  • Makes it easier to handle structured POJO exchanges with LLMs.
• Workarounds exist (manual TOON serialization), but a built-in module would provide first-class support aligned with Spring AI’s extensibility model.