{"id":"wax","name":"Wax","homepage":"https://github.com/christopherkarani/Wax","repo_url":"https://github.com/christopherkarani/Wax","category":"memory-storage","subcategories":["rag","on-device-ai","ios-macos","vector-search"],"tags":["swift","rag","vector-search","apple-silicon","metal","bm25","hnsw","on-device","ios","macos","memory","mcp"],"what_it_does":"A sub-millisecond on-device RAG memory library for iOS and macOS AI agents, implemented in pure Swift with hybrid BM25 + HNSW vector search, local MiniLM embeddings, Metal GPU acceleration, token budgeting, and a single portable .wax file format — no server or API required.","use_cases":["Persistent, searchable memory for iOS and macOS AI agent apps without cloud dependencies","Privacy-first RAG where all embeddings and retrieval happen on-device","Token-budget-aware context compression for constrained LLM context windows","Embedding Apple Silicon GPU acceleration into Swift AI applications via Metal"],"not_for":["Cross-platform or server-side RAG systems (Swift/Apple platform only)","Large-scale document corpora exceeding single-device storage","Teams not working in Swift or the Apple ecosystem"],"best_when":"You are building a native iOS or macOS app that needs fast, private, on-device memory/RAG without any network calls or server infrastructure.","avoid_when":"You need a cross-platform, server-hosted, or cloud-scalable vector database — use Pinecone, Weaviate, Qdrant, or Chroma instead.","alternatives":["chromadb","qdrant","pinecone","sqlite-vec","mlx-embeddings"],"af_score":78.9,"security_score":80.0,"reliability_score":null,"package_type":"mcp_server","discovery_source":["github"],"priority":"low","status":"evaluated","version_evaluated":"0.1.8","last_evaluated":"2026-03-01T09:50:06.391369+00:00","performance":{"latency_p50_ms":1,"latency_p99_ms":null,"uptime_sla_percent":null,"rate_limits":null,"data_source":"vendor_claimed","measured_on":null}}