zetic_mlange 1.10.0

Flutter FFI SDK for running ZeticMLange on-device AI models on Android and iOS.

Unreleased

• Removed unsupported general-model target and AP type selectors.
• Added Hugging Face cache handling policy configuration.
• Added synchronous LLM tool calling with runWithTools() token streaming.

1.9.1

Flutter v1.9.1.

• Aligned Flutter RAG APIs with Android and iOS by moving RAG generation to RagPipeline.
• Removed model-level RAG helper methods from ZeticMLangeLLMModel.
• Added runtime-key model and LLM creation with exact typed activation errors, including sdk_upgrade_required for older native SDKs.
• Added immutable runtime activation metadata (activationId, completion state, and receipt status) for runtime-key model and LLM creation without exposing receipt tokens.
• Added release-set verification for canonical header pins and required native activation-result symbols.

1.9.0

Flutter v1.9.0.

• Made Flutter model lifecycle closes idempotent and guarded post-close FFI calls.
• Aligned LLM lifecycle naming around close() while keeping deprecated deinit() compatibility.
• Added LFM-VL image+text runtime APIs for supported local vision-language models.
• Added function calling APIs for LLM tool-use workflows.
• Added on-device RAG APIs for retrieval and context-grounded generation.
• Added KV state persistence APIs for saving, loading, and resetting LLM state.

1.8.1

Flutter v1.8.1.

• Fixed Android Gradle configuration for published pub.dev packages by removing the plugin-local local.properties lookup.

1.8.0

Flutter v1.8.0.

• Initial pub.dev-ready release of the ZeticMLange Flutter SDK.
• Added Dart API wrappers for ZeticMLange general, Hugging Face, and LLM models.
• Added tensor-based inference APIs for Android and iOS.
• Added LLM prompt execution and token streaming APIs.
• Added model mode, quantization, target, accelerator, and cache policy options.