# sonos/tract

Tiny, no-nonsense, self-contained, Tensorflow and ONNX inference

Repository: https://github.com/sonos/tract
Canonical: https://ross.abutalabs.com/products/tract
Language: Rust
License: NOASSERTION
License Family: other
Topics: tensorflow, onnx, neural-networks, artificial-intelligence, rust-library, rust
Last push: 2026-08-26T15:28:47+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 98, longevity 100
- inputs: {"age_days": 3313, "days_push": 7, "days_rel": 14, "gap_med": 20.0, "n_releases_24m": 21}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3045, forks 281 (observed 2026-08-28T04:07:39.504806+00:00)

## What it is
Tract is Sonos' tiny, self-contained neural-network inference engine written in Rust. It loads ONNX, TensorFlow/TFLite, and NNEF models, optimizes them, and runs them on everything from embedded ARM CPUs to GPUs and WebAssembly.

## Use cases
- run onnx models in rust
- neural network inference on embedded arm devices
- run llm inference locally in rust
- speech recognition inference for wake word detection
- run stable diffusion models without python
- run machine learning models in the browser via webassembly
- deploy tensorflow lite models on edge devices

## When to choose
- you need a small, self-contained inference runtime with no heavy dependencies
- you target embedded or resource-constrained hardware
- you want to ship a tiny runtime with models pre-translated to NNEF
- you need ONNX or TFLite inference in Rust or Python across many platforms

## When to avoid
- you need the full training or ecosystem tooling of PyTorch or TensorFlow
- you require broad operator coverage for exotic models
- you depend on internal tract crates, which are not a stable API

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, speech-recognition, image-processing, stable-diffusion
- domain: machine-learning, deep-learning, artificial-intelligence, speech-processing, computer-vision
- platform: rust, python, windows, wasm, embedded, cross-platform
- tags: onnx, nnef, tensorflow, tflite, inference-engine, embedded-ai, neural-networks, linux, macos, gpu

## Member repositories
- sonos/tract (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:39.504806+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 2026-08-30T07:28:58.010225+00:00, confidence not recorded.
  - readme: https://github.com/sonos/tract (fetched 2026-08-28T04:07:39.504806+00:00, sha 652f2838e9fe)
  - registry_crates: https://crates.io/api/v1/crates/tract (fetched 2026-08-29T09:44:08.194413+00:00, sha b66d0e979ccc)
- Data as of 2026-08-30T08:39:29.467469+00:00.