ai_edge 0.2.0

Flutter plugin for on-device AI inference powered by MediaPipe GenAI. Run large language models directly on iOS and Android with optimized performance.

0.2.0

Breaking Changes

• API Simplification: Replaced config objects with individual named parameters (#11)
  • createModel() now accepts individual parameters instead of ModelConfig object
  • createSession() now accepts individual parameters instead of SessionConfig object
  • initialize() now accepts individual parameters for both model and session configuration
  • Config classes (ModelConfig and SessionConfig) are still available internally but not exposed in public API

Improvements

• Developer Experience:

  • More intuitive API with direct parameter passing
  • Better IDE autocomplete support without config object nesting
  • Optional parameters with sensible defaults:
    • maxTokens: Default 1024 (previously required)
    • temperature: Default 0.8
    • randomSeed: Default 1
    • topK: Default 40

• Platform Implementation:

  • Made non-critical parameters optional in Android native implementation
  • Fixed platform channel communication issues with Unit type handling

• Code Quality:

  • Removed unnecessary toString() override from GenerationEvent class
  • Updated documentation to remove references to deprecated exception types
  • Updated tests to reflect the new API design

0.1.0

Breaking Changes

• Package Structure: Reorganized package to use src/ directory structure

  • Main implementation moved to lib/src/ directory
  • Public API now exported through lib/ai_edge.dart
  • Platform interfaces moved to src/ subdirectory

• Error Handling: Complete overhaul of error handling system

  • Introduced sealed classes (AiEdgeException) for exhaustive error handling
  • Replaced string-based errors with type-safe exception hierarchy
  • CallError now accepts Exception directly instead of message strings
  • Removed unnecessary PlatformException conversions

• Platform Interface: Standardized data conversion patterns

  • Platform interfaces now accept Map<String, dynamic> instead of typed objects
  • Data conversion happens in upper layer (ai_edge.dart) for consistency
  • Cleaner separation between platform communication and business logic

Features

• Monorepo Structure: Project reorganized as Flutter workspace
  • Packages moved to packages/ directory
  • Examples moved to examples/ directory
  • Enables better package management and development workflow

Improvements

• Code Quality

  • Improved test coverage with updated mock implementations
  • Better separation of concerns with src/ directory structure
  • More maintainable and extensible codebase

• Developer Experience

  • Cleaner public API surface
  • Better IDE support with organized file structure
  • Improved error messages with sealed class pattern

Chores

• Added LICENSE and README.md symlinks for pub.dev publishing requirements
• Updated dependencies and build configurations
• Improved Android build configuration

0.0.1

Initial release of the AI Edge Flutter plugin for on-device AI inference using MediaPipe GenAI.

Features

• Native AI Inference: On-device AI inference support for both iOS and Android platforms using MediaPipe GenAI
• Streaming Responses: Real-time streaming text generation with async stream support
• Model Management: Built-in model download and management capabilities
• Example App: Comprehensive example application with chat UI and model management interface
• Platform Support:
  • iOS 13.0+ with MediaPipeTasksGenAI (0.10.24)
  • Android with MediaPipe GenAI (0.10.24)

Infrastructure

• Comprehensive CI/CD workflow with platform-specific builds
• Integration and unit test infrastructure
• Automated code formatting and linting
• Pull request template for consistent contributions

Developer Experience

• Project documentation for pub.dev release
• Tool version management with mise configuration
• Simplified plugin architecture with clean API design
• Example app demonstrating all plugin capabilities