# Run models on your machine with the local provider

> An experimental local provider runs a model inside the Yolop binary, with Metal and CUDA release builds and weights fetched by yolop weights pull.

Source: <https://everruns.com/docs/changelog/2026-08-21-yolop-local-inference-provider/>

Date: 2026-08-21
Products: Yolop
Category: Feature
Available: In Yolop 0.17.1
Pull request: everruns/yolop#581
Pull request: everruns/yolop#599
Pull request: everruns/yolop#623

`--provider local` runs a model inside the Yolop process: no server, no API key and no network during a turn. The default model is Qwen3-30B-A3B-Instruct in Q4 GGUF. Fetch the weights first, since a turn never waits on a download:

```bash
yolop weights pull
yolop --provider local
```

The provider is experimental. The engine ships in separate release archives, `yolop-<target>-metal.tar.gz` and `yolop-x86_64-unknown-linux-gnu-cuda.tar.gz`. The default archive, which Homebrew installs, leaves it out, so the Linux tarball dropped from 29.9 MB to 18.2 MB. From source, build with the `metal` or `cuda` feature.

See [Providers and models](https://github.com/everruns/yolop#providers-and-models), [#581](https://github.com/everruns/yolop/pull/581), [#599](https://github.com/everruns/yolop/pull/599) and [#623](https://github.com/everruns/yolop/pull/623).