on_device_ai 0.2.1

A cross-platform Flutter plugin for on-device AI/ML inference. Supports text sentiment analysis and image classification using Apple frameworks (iOS) and TensorFlow Lite (Android).

0.2.1

• Fixed pub.dev platform score warning by relocating Package.swift to ios/on_device_ai/Package.swift and macos/on_device_ai/Package.swift to strictly follow Flutter's Swift Package Manager structure guidelines.

0.2.0

• Added macOS platform support (macOS 11.0+) utilizing NSImage and native Apple frameworks (AppKit, Vision, NaturalLanguage).
• Updated example app to include a macOS runner with file picking support.

0.1.2

iOS Improvements

• Added Swift Package Manager (SPM) support via ios/Package.swift. The plugin now works in Xcode projects that use SPM instead of CocoaPods, resolving the pub.dev platform score warning.
• Updated on_device_ai.podspec with correct version number (0.1.2), accurate description, and explicit declarations for the NaturalLanguage, Vision, and CoreML system frameworks.

---

0.1.1

Documentation

• Added full dartdoc comments to all public API elements: AIResult, ModelConfig, OnDeviceAi, and OnDeviceAiPlatform. This satisfies the pub.dev documentation score requirement (≥20% of public API elements documented).

---

0.1.0

New Features

• Image Classification (runImage): Pass an image byte array and receive an on-device classification result (e.g. "Golden Retriever: 94%") with no internet connection.
  • Android: Uses a bundled MobileNet V1 quantized TFLite model (~4MB) via the stable base tensorflow-lite:2.11.0 interpreter.
  • iOS: Uses Apple's built-in Vision framework (VNClassifyImageRequest) — zero model downloads required.
• Multimodal Example App: The example app now includes an image picker UI to test both text and image AI pipelines on the same screen.

Improvements

• Android threading: All TFLite operations (loadModel, runText, runImage, dispose) now run on a background Dispatchers.IO coroutine to prevent UI-thread blocking.
• iOS sentiment reliability: Fixed a bug where NLTagger always returned "Neutral" by switching from tag(at:) to enumerateTags and explicitly forcing the .english language tag.

Bug Fixes

• Fixed a native JNI crash (SIGSEGV in libtask_vision_jni.so) on Android 10 devices caused by tensorflow-lite-task-vision:0.4.4. Replaced with a stable manual inference pipeline using tensorflow-lite:2.11.0.
• Fixed a Gradle build failure (Plugin [id: 'com.android.library'] was not found) by migrating the Android plugin build script from Kotlin DSL (build.gradle.kts) to Groovy (build.gradle).

---

0.0.1

• Initial release.
• Support for on-device Text Inference via Core ML (iOS) and TensorFlow Lite (Android).
• Unified Dart API for model loading, text inference, and streaming.