Document the planned managed model setup: one recommended speech bundle, explicit installation, offline reuse, a simpler API/example and native build packaging requirements. Runtime behavior is unchanged.

Replace the generic library skeleton with flutter_local_voice_agent.

Add local model integrity validation and staged installation, typed events, lifecycle controls, and bounded Dart event delivery.

Add native sherpa-onnx 1.12.14 VAD, streaming transducer recognition and VITS synthesis integration with generation-based output cancellation.

Add Android/iOS foreground microphone, playback and interruption bridges.

Add a local-model example, reproducible native provisioning and test tools.

Keep physical-device, acoustic, thermal, voice quality and redistribution gates explicit. This is not a production-qualified or published release.

Add optional bounded local llama.cpp integration and regression coverage for typed lifecycle failures, manifest roles and UTF-8 cancellation handling.

Add GitHub Actions checks, patch-release tagging, and OIDC publication workflow.