Circuitry Logo
Circuitry
Open Circuitry

OpenClaw Outside Circuitry: the Circuit Plugin

For OpenClaw running on its own: use the models your Circuit shares, including those on every computer in your cluster.

Using OpenClaw inside Circuitry? You don't need this. Circuitry gives it your Circuit's models automatically: install Circuit on each computer and put them in a cluster. See OpenClaw in Circuitry.

How you run OpenClawWhat you need
Inside CircuitryCircuit on each computer, in a cluster. No plugin
On its own, such as from a terminal or TelegramThis plugin, paired once with your Circuit

Install the Circuit plugin in OpenClaw and pair it once. If your Mac and your PC are in a cluster, the models on both appear in OpenClaw under one provider.

What you need

  • OpenClaw 2026.10.9 or newer.
  • Circuit on the computer you pair with, up to date.
  • Both computers on the same network.

To reach your whole cluster, pair with the lead Circuit: the one you add machines from under Cluster… in the Circuit menu. A computer in a cluster only knows its own models. See Compute Clusters.

Install the plugin

On the computer running OpenClaw:

openclaw plugins install git:github.com/johnwylie70/openclaw-circuit

Pair with your Circuit

openclaw models auth login --provider circuit
  1. Type your Circuit's address, the IP Address in its menu, such as 192.168.1.20.
  2. A window opens on that computer, naming the app and the computer asking, with a 6-digit code.
  3. Type the code into OpenClaw. If you did not start this, press Not me in the window.

The window appears only when pairing. After that, OpenClaw connects without asking.

Then run openclaw gateway restart so OpenClaw picks up the models.

Choose a model

openclaw models list --provider circuit --refresh
openclaw models set circuit/gemma4:latest

A model on the computer running your Circuit is circuit/<model>. A model on another computer in the cluster is circuit/<computer>/<model>, such as circuit/pc/qwen3.8:27b.

What OpenClaw can reach

OpenClaw can list the shared models and chat with them. It cannot reach your files, terminals or code, and it cannot run anything on your computers.

To remove its access, choose Cluster… in the Circuit menu and press Remove next to OpenClaw under Apps using these models. It stops working at once.

When a computer is short of memory

The computer holding the model decides what happens:

  • The model isn't running: it starts, and the first reply takes longer.
  • Other models are in the way: they stop to make room. Circuitry shows a message saying which, so you can adjust.
  • It can't make enough room: nothing stops, and OpenClaw is told how much the model needs and what is using the memory.

To try another computer when that happens, list a fallback in OpenClaw:

// openclaw.json
{
  agents: {
    defaults: {
      model: {
        primary: "circuit/Qwen3.8-27B-4bit",
        fallbacks: ["circuit/pc/qwen3.8:27b"]
      }
    }
  }
}

Fallbacks apply to your default model. A model you pick for one chat with /model runs as chosen, and OpenClaw tells you if it can't.

Good to know

  • Text models only, for now.
  • Your computers have to be on the same network. Home Circuit isn't supported.
  • OpenClaw reaches your cluster through the lead Circuit, so a model on your PC stops answering while the lead computer sleeps, even with OpenClaw on that same PC.
  • A model that isn't loaded can't report its context, so it may show as 64k until it is loaded. The plugin's page shows how to set the size yourself.

Related