Optional TOON support for token-efficient structured context?

[Julian Mayer (ips-jm)]

I’m looking at TOON (Token-Oriented Object Notation) as a token-efficient, lossless alternative to JSON for structured payloads sent to LLMs:
https://github.com/toon-format/toon#readme

Agent Framework often feeds structured data into prompts (tool outputs, RAG results, memory/state). JSON can get pretty token-heavy, and TOON seems designed exactly for this boundary (nested structure + tabular arrays for uniform objects).

Question:

• Would you be open to an optional TOON formatter/serializer (JSON remains default) to reduce tokens/cost for large structured context?
• Where would you see the best integration point in Agent Framework (middleware hook, tool result formatting, memory/context serializer)?

Refs:

• TOON Spec: https://github.com/toon-format/toon
• TOON .NET: https://github.com/toon-format/toon-dotnet
• TOON Python: https://github.com/toon-format/toon-python