# dblalock/bolt

10x faster matrix and vector operations

Repository: https://github.com/dblalock/bolt
Canonical: https://ross.abutalabs.com/products/dblalock-bolt
Language: C++
License: MPL-2.0
License Family: copyleft
Topics: machine-learning, data-mining, database, compression
Last push: 2022-10-12T21:47:45+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 3500, "days_push": 1421, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2514, forks 174 (observed 2026-08-28T04:06:57.536337+00:00)

## What it is
Bolt is a C++ library (with a Python wrapper) for lossy compression of dense real-valued vectors that supports mathematical operations directly on compressed representations, offering 10-200x space and compute savings. It also includes MADDNESS (mithral), an approximate matrix multiplication algorithm aimed at accelerating neural network inference on CPUs.

## Use cases
- compress large collections of dense vectors with minimal accuracy loss
- speed up dot products and distance computations on compressed vectors
- accelerate neural network inference on CPUs with approximate matrix multiplication
- reduce memory footprint of embedding or feature vectors
- research approximate matrix multiplication algorithms

## When to choose
- you have mostly-dense vectors and can tolerate lossy compression
- CPU inference speed is a bottleneck and 10-200x approximation savings are acceptable
- you need theoretical error guarantees on approximations

## When to avoid
- you need exact results
- you require GPU support or convolutions for MADDNESS
- you need a well-maintained Python wrapper, since it reportedly no longer builds for many users

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, compression, math, benchmarking
- domain: machine-learning, databases, performance
- platform: cpp, python, cross-platform
- tags: vector-compression, approximate-matrix-multiplication, maddness, dense-vectors, lossy-compression, simd, nearest-neighbor, algorithms, linux, macos

## Member repositories
- dblalock/bolt (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:57.536337+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:26:13.480757+00:00, confidence not recorded.
  - readme: https://github.com/dblalock/bolt (fetched 2026-08-28T04:06:57.536337+00:00, sha 1742c86809f9)
- Data as of 2026-08-30T08:39:29.467469+00:00.