Edge – Vector Database in Zig Using USearch and RocksDB (github.com)

Antarys Edge is a lightweight vector database implementation written in Zig (targeting Zig 0.15.1) that combines Usearch (HNSW indexing/search) with a RocksDB fork and http.zig to provide a small-to-medium scale, Zig-native vector store. It ships with compatible Python/Node client APIs for vector operations (embeddings planned) and is intended as a drop-in starting point for teams experimenting with the Antarys client API or building a custom Zig-based vector DB. Repo: github.com/antarys-ai/edge with simple zig build/run instructions; currently tested on Apple ARM only and known to have UBSan/linker issues in production builds because of the RocksDB dependency. Technically, Edge uses Usearch for nearest-neighbor (HNSW) search and RocksDB for storage, exposes search controls (expansion, thread count, limits), supports per-result filtering, optional vector retrieval, and maps internal keys to external IDs via an IdMap. Example code shows careful error mapping, allocator-managed result arrays, and concurrency utilities (ThreadPool, BatchCoordinator, work-stealing) for parallel tasks. Search latency can approach Qdrant/Pinecone, but indexing is still slower — so it’s best suited for prototyping, embedding into Zig projects, or custom extensions rather than high-throughput production. Platform support, production build stability, and embedding ingestion are the main areas to watch as the project matures.