ondevice-rag-engine

What problem does it solve?

This Skill provides a complete on-device retrieval-augmented generation (RAG) pipeline, enabling semantic search, embedding generation, chunking, indexing, and vector similarity without sending data to external services.

Core Features & Use Cases

• On-device embeddings: Generate sentence embeddings using Apple's NaturalLanguage framework for private data.
• Local vector store: Store vectors in SwiftData with deterministic and queryable indexing.
• Fast similarity search: Use cosine similarity powered by Accelerate for efficient retrieval.
• Chunking strategies: Break meetings and notes into context-preserving chunks for accurate matching.
• Use case: Planner apps can semantically search meetings, notes, and decisions entirely offline.

Quick Start

Index your first MeetingRecord by chunking content and storing embeddings locally, then run a query like "What were the key decisions from the last meeting?" to retrieve relevant chunks. Ensure embedding model is available on-device and do not perform any network calls.