# trymirai/uzu

A high-performance inference engine for AI models

Repository: https://github.com/trymirai/uzu
Canonical: https://ross.abutalabs.com/products/uzu
Homepage: https://trymirai.com
Language: Rust
License: MIT
License Family: permissive
Topics: high-performance, inference, llm, metal, rust, ai, tts
Last push: 2026-08-26T17:19:59+00:00

## Health v2 (maintenance only)
Score: 85/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 31
- inputs: {"age_days": 436, "days_push": 7, "days_rel": 7, "gap_med": 1.0, "n_releases_24m": 19}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1678, forks 73 (observed 2026-08-28T04:05:20.975997+00:00)

## What it is
Uzu is a high-performance inference engine written in Rust for running AI models directly on-device, with Python, TypeScript, and Swift bindings. It leverages Apple Silicon's unified memory and Metal to deliver zero-latency, private, cost-free inference inside applications.

## Use cases
- run llm inference locally on apple silicon
- embed an ai model directly in an ios or macos app
- serve a local api for chatting with models
- generate structured output from a local model
- run text-to-speech on device
- avoid cloud inference costs and keep data private
- deploy ai in an app with zero latency

## When to choose
- you target Apple Silicon (macOS/iOS) and want fast on-device inference
- you need full data privacy with no data leaving the device
- you want to eliminate per-token cloud inference costs
- you want simple high-level APIs across Rust, Python, TypeScript, or Swift
- you need low-latency, interactive AI experiences in your app

## When to avoid
- you need to run on non-Apple hardware such as Linux, Windows, Android, or CUDA GPUs
- you need broad cross-platform model support beyond the curated model list
- you require serving many concurrent users with batched throughput rather than batch size 1
- you need a mature ecosystem like llama.cpp or vLLM with extensive community tooling

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, sdk, tts, machine-learning
- domain: large-language-models, machine-learning, artificial-intelligence, cross-platform
- platform: rust, python
- tags: on-device-ai, metal, apple-silicon, local-inference, quantization, unified-memory, edge-ai, privacy-preserving, structured-output, chat-completions, macos, ios, swift, nodejs, gpu

## Member repositories
- trymirai/uzu (main) score 85

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:20.975997+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 2026-08-30T03:41:35.060699+00:00, confidence not recorded.
  - readme: https://github.com/trymirai/uzu (fetched 2026-08-28T04:05:20.975997+00:00, sha ac5caabfed1a)
  - homepage: https://trymirai.com (fetched 2026-08-29T11:15:01.149001+00:00, sha 2fedc1aa546d)
  - site_page: https://docs.trymirai.com/ (fetched 2026-08-29T11:15:01.151592+00:00, sha 298eca380341)
  - site_page: https://trymirai.com/about-us (fetched 2026-08-29T11:15:01.153265+00:00, sha dec7cd08ce25)
- Data as of 2026-08-30T08:39:29.467469+00:00.