Android Local RAG & Vector Databases 🔎

Retrieval-Augmented Generation (RAG) on Edge devices provides privacy-first, low-latency contextual intelligence. Instead of using cloud Pinecone/Weaviate, this skill enforces local vector embeddings via Room or SQLite.

⚡ When to Use

• When tasked with "RAG", "semantic search", or "vector embeddings".
• Building highly private AI features.

🏗️ Core Rules / Pillars

1. Vector Search in Room

• Pattern: Standard Room does not have native vector similarity out of the box in stable. Use L2 distance extensions or sqlite-vss.
• Implementation: Avoid pulling unmaintained third-party vector DBs. Use SQLite virtual tables with FTS or simple cosine similarity functions if embeddings are small.

🚧 Critical Anti-Hallucination Guards

• Trap: Do NOT hallucinate that Room natively supports SELECT * FROM tbl ORDER BY VECTOR_DISTANCE() securely out of the box without specific extensions.