Back to browse
GitHub Repository

High-performance, Hardware-First Vector Database engine built in Zig and Go. Sub-millisecond HNSW search with 64-bit metadata filtering and 1.8MB binary footprint.

19 starsZig

DeraineDB – A 33MB Vector DB in Zig/Go with Sub-Millisecond HNSW

by RikardoB·Mar 5, 2026·4 points·1 comment

AI Analysis

●●SolidWizardryBig Brain

Sub-millisecond HNSW in 33MB, but Milvus and Weaviate already own this space.

Strengths
  • Memory-mapped payload segregation with strict cache-line alignment prevents corruption cleanly
  • Zero-copy CGO bridge via unsafe.Pointer avoids GC pauses on large vector batches
  • Bitwise metadata filtering in HNSW hot loop resolves categories in single CPU cycle
Weaknesses
  • Crowded vector DB category with well-funded competitors (Pinecone, Milvus, Weaviate)
  • No evidence of production adoption or real-world performance vs. established tools
Target Audience

Edge computing engineers, local AI/RAG developers, embedded systems builders

Similar To

Milvus · Weaviate · Qdrant

Similar Projects

CLI tool to analyze your Vector Embeddings!

Embedding auditor with 5 checks and pretty plots, but crowded niche with unclear novelty.

Ship It
gauravvij137
213mo ago