Wax is a Swift library providing a memory layer for on-device AI agents. It’s described as “The SQLite for AI memory. One file. Full RAG. Zero infrastructure.”

Created by Christopher Karani, Wax replaces complex RAG pipelines with a single portable file format, designed for privacy-first, offline-capable AI applications on Apple platforms.

Documents

DocumentDescription
Architecture and StorageTechnical deep-dive into Wax’s custom binary format, SQLite FTS5 integration, HNSW vector indexing, and Metal GPU acceleration
  • SQLite - Wax uses SQLite FTS5 for BM25 full-text search
  • Vector Search - HNSW implementation via USearch library