Skip to content
AnyRouter
Local relay · Live

Your Mac is now
an AI backend.

Serve Apple Foundation Models — or any local model — through the same API as 500+ cloud models.

The relay bridges AnyRouter's gateway to hardware you own. Your device dials out, holds an encrypted connection, and answers inference requests with the model running on your desk — streaming, revocable, and billed at exactly $0.

Pair a deviceRead the docs
Your appone OpenAI-compatible API, same key, same SDK
AnyRouterroutes, streams, and fails over between cloud and your devices
Your MacApple Foundation Models, Ollama, LM Studio…
01

Setup

How it works

STEP 1

Pair your device

Register your Mac from the dashboard and get a one-time pairing token. Revoke it any time — the device is yours, the trust is scoped.

STEP 2

Start the relay

Run fm serve (or Ollama / LM Studio), then one CLI command. It auto-detects your local server and dials out to AnyRouter — no port forwarding, no public IP, no inbound firewall holes.

fm serve
anyr relay start --pool
STEP 3

Route as usual

Call the same OpenAI-compatible API you already use. Requests stream down the encrypted bridge to your device, and the response streams back — token by token.

02

Capabilities

Built for hardware you own

Apple Foundation Models first

Serve the on-device Apple Intelligence model straight from your Apple Silicon Mac via Apple's own local server. Private, fast, and entirely on your hardware.

Any local model works

The relay speaks to any OpenAI-compatible local server — Ollama, LM Studio, llama.cpp, vLLM. If it runs on your machine, it can serve through AnyRouter.

Outbound-only, revocable

Your device opens the connection; the cloud never reaches in. Pairing tokens are hashed at rest and revocable from the dashboard with one click.

Fails soft, never hangs

Mac asleep? Laptop in a bag? Requests fail over to your other configured upstreams in seconds — the gateway never stalls waiting on an offline device.

$0 routing

It's your compute and your model — relay-served requests are never billed against your credits, the same policy as BYOK.

Live device dashboard

See every paired device, its live connection status, and connects-over-time analytics in the dashboard.

03

Roadmap

Where this is going

The relay is a two-way street: today your device serves you; next, your device can serve the network.

Serve Apple Foundation Models

Live

Pair a Mac, run fm serve + the relay, and route to the on-device Apple Intelligence model through the standard API as apple/foundation-model.

Any local LLM behind the same bridge

Live

Point the relay at Ollama, LM Studio, or any OpenAI-compatible localhost server and expose it as your own private upstream.

Cloud-to-client service mesh

Live

The bridge now runs both ways: run anyr relay start --pool to donate your device's idle capacity, and other users' matching requests route to it when your own capacity is idle. Donate a valid device and get Go, same as key donation.

Share coding-agent capacity to the pool

Planned

Proxy locally-installed coding agents and CLI subscriptions (Antigravity and other agent CLIs) through the relay and donate idle capacity to the shared pool — the same way key donation works today, credits included.

Idle compute is the next donated key

Today you can donate provider keys to the shared pool and get Go. The relay extends that to compute: share a local model or an idle coding-agent seat and get Go the same way.

See how key donation works
04

FAQ

Frequently asked questions

Put your hardware on the network

Pair a device now — you'll be ready the moment the relay CLI ships, and first in line for the compute pool.

Pair a deviceCreate an account
See also:Bring your own keys (BYOK)Donate keys, get GoFree tier