akashi_gemma 0.3.0

On-device LanguageModel for the Akashi agent framework, backed by flutter_gemma: run agents fully offline, or as the primary in a cloud FallbackModel chain.

Changelog

0.3.0

• Initial release of the on-device adapter.
• GemmaModel implements LanguageModel over a GemmaBackend seam, normalizing on-device tokens → TextDeltaPart, thinking → ReasoningDeltaPart, and function calls → ToolCallCompletePart. No HTTP — proves the contract is not transport-bound. Pairs with akashi_gateway's FallbackModel. Streaming honors request.cancel to stop long on-device generation.
• FlutterGemmaBackend wraps flutter_gemma 1.x's InferenceChat for real device inference (including parallel function calls); the GemmaBackend seam keeps normalization testable offline. Requires Dart ≥3.12 / Flutter ≥3.44.
• GemmaProvider mints GemmaModels over a shared backend.