Install on your Macs and they find each other automatically via mDNS. No IP addresses to type, no config files to edit.
Trust-on-first-use pairing with Ed25519 identity keys. Both sides must accept before any inference flows between peers.
Native macOS menu bar app with Metal-accelerated inference via MLX. Runs on Apple Silicon with no CUDA, no Docker.
Drop-in replacement for the OpenAI API. Point any client, IDE extension, or tool at localhost:7258 and go.
Large models are automatically split across the GPUs in your cluster. Each Mac handles a shard of the model layers.
Everything stays on your local network. No cloud, no telemetry, no API keys to manage. Completely self-contained.
Download the app or install via pip. Each machine gets a unique Ed25519 identity on first launch.
Salut advertises via mDNS on your local network. Peers appear in each other's menu bar within seconds.
You get a notification when a new peer wants to connect. Accept it, and they accept yours — mutual trust established.
Send a request to any peer's API. If the model is too large for one machine, Salut coordinates the cluster to run it together.
