ℹ️ I see that Google's recommended solution for LLM inference 'on the Edge' has become
LiteRT-LM (with MediaPipe now
deprecated for LLM).
This is surely an important integration for
Koog.ai, as it supports multiple platforms (while not having a KMP wrapper yet).
However, the project doesn't yet have support for structured output via its Kotlin bindings!
⚠️
This is a critical weakness - for the broadest range of use-cases require structured output; only toy projects or narrow language-only cases can get away without it.
There's an issue here to
request Kotlin structured output be exposed.