ColdIVF

ColdIVF is a read-optimized vector index in Rust that builds an IVF index in one file for cold reads.

ColdIVF is a read-optimized vector index implemented in Rust and built for cold reads. It stores an IVF index in a single file by splitting vectors into clusters and gathering nearby clusters into groups. The system keeps a compact 1-bit code of each vector inside its cluster alongside a more precise copy. A cold query reads only the necessary clusters and candidate copies in two batches, avoiding loading the entire index into memory. It supports Python via coldivf-python and provides methods for compaction and calibration. The platform is designed for developers who need efficient vector search performance with low memory usage during builds.

ColdIVF at a glance

Best for: Software engineer
Runs on: macOS, Linux, Python, Rust
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/13lack13lood/coldivf

Key features

Searches based on 1-bit codes
Batched disk reads via FRDADVISE and iouring
Cluster ordering for fewer reads
Small routing tables with 4-bit values
SIMD scan supporting NEON, AVX2, and AVX-512
Page-aligned rows
Bounded build memory
Background maintenance and compaction

Use it for

Building vector search indexes with low cold-read overhead
Managing vector embeddings in Python and Rust applications
Executing fast similarity searches reading under 1 percent of data
Handling writes and maintenance concurrently with background reads

Alternatives to ColdIVF

TurboVec — Fits a 10 million document corpus in 4 GB of RAM and searches faster than FAISS.
zvec — It is a pure local, serverless vector database providing fast hybrid search.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
HelixDB — Combines graph, vector, and full-text search in one engine.
RuVector — Provides persistent vector retrieval, graph relationships, and local decision making in Rust.
Stof — Runs AI-generated logic safely in a sandbox inside the app.
ort — Provides an ergonomic Rust API for running high-performance ONNX models.
Reflex — Compiles CUDA kernels ahead of time for fast cold starts without runtime JIT.
USearch — It uses SIMD instructions and optimized concurrency to outperform FAISS.
zerostack — Minimal memory footprint and high CPU efficiency compared to JS-based agents.
Shimmy — Runs local AI models with zero Python or C++ dependencies.
fal — Access 1,000+ generative models and high-performance serverless GPUs.

All ColdIVF alternatives