ai_edge_rag 0.0.1

ai_edge_rag: ^0.0.1 copied to clipboard

Features #

• RAG Support: Complete implementation of Retrieval Augmented Generation capabilities

  • Semantic search using vector similarity
  • Context-aware response generation
  • Document retrieval from vector stores

• Embedding Models: Support for both local and cloud-based embeddings

  • Local embedding models (Gemma and Gecko)
  • Gemini API-based embeddings for cloud processing
  • On-device embedding generation for privacy

• Vector Storage: Flexible storage options for embeddings

  • In-memory vector store for fast access
  • SQLite vector store for persistent storage
  • Automatic persistence across app restarts

• Document Processing: Comprehensive text chunking capabilities

  • Manual chunk management (memorizeChunk, memorizeChunks)
  • Automatic text chunking with configurable size and overlap
  • Support for large document processing

• Streaming Responses: Real-time text generation

  • Streaming API for progressive response display
  • Full text accumulation in partialResult
  • Event-based completion notification

• Customizable Retrieval: Fine-tuned context retrieval

  • Configurable topK for number of retrieved chunks
  • minSimilarityScore threshold for relevance filtering
  • System instructions for RAG behavior control

Documentation #

• Comprehensive README.md with usage examples
• Detailed Dart doc comments for all public APIs
• API reference documentation
• Best practices and troubleshooting guides
• Use case examples and implementation patterns

Testing #

• Complete unit test suite (28 tests)
  • Model and session creation tests
  • Embedding model configuration tests
  • Document memorization tests
  • RAG generation tests
  • Type validation tests

Platform Support #

• Android support (API level 24+)
  • GPU acceleration support
  • Hardware backend selection
  • Large heap configuration for big models

API #

Core classes and methods:

• AiEdgeRag - Main entry point
• createEmbeddingModel() - Local embedding model setup
• createGeminiEmbedder() - Gemini API embedder setup
• memorizeChunk() - Single chunk storage
• memorizeChunks() - Batch chunk storage
• memorizeChunkedText() - Automatic text chunking and storage
• setSystemInstruction() - RAG behavior configuration
• generateResponseAsync() - Context-aware streaming generation
• EmbeddingModelConfig - Local embedding configuration
• GeminiEmbedderConfig - Gemini API configuration
• SystemInstruction - RAG system instruction
• VectorStore - Storage type enum (inMemory, sqlite)
• EmbeddingModelType - Model type enum (gemma, gecko)

Known Limitations #

• iOS support not yet implemented (Android only)
• Embedding models must be in MediaPipe format
• SQLite vector store uses basic similarity search without advanced indexing