You don’t add the Mac mini as “an agent machine” directly. You add its Qwen server as a model provider, then optionally create an OpenClaw agent that uses that model.

Best setup:

1. On the Mac mini: expose Qwen via OpenAI-compatible API

If using Ollama:

OLLAMA_HOST=0.0.0.0:11434 ollama serve

Check model name:

ollama list

From this machine, test:

curl http://MAC_MINI_IP:11434/v1/models

Use /v1 in the OpenClaw config.


2. Add a provider to OpenClaw

In ~/.openclaw/openclaw.json, add something like:

{
  models: {
    providers: {
      "mac-qwen": {
        baseUrl: "http://MAC_MINI_IP:11434/v1",
        api: "openai-completions",
        apiKey: "ollama",
        request: {
          allowPrivateNetwork: true
        },
        models: [
          {
            id: "qwen3.5",
            name: "Qwen 3.5 on Mac mini",
            reasoning: true,
            input: ["text"],
            contextWindow: 32768,
            contextTokens: 24000,
            maxTokens: 8192,
            cost: {
              input: 0,
              output: 0,
              cacheRead: 0,
              cacheWrite: 0
            }
          }
        ]
      }
    }
  }
}

Important bit: because it’s on your LAN, OpenClaw needs:

request: { allowPrivateNetwork: true }

Otherwise the provider fetch guard may block private-network model endpoints.


3. Add it to the allowed model list

Under:

1
2
3
4
5
6
7
8
9
agents: {
  defaults: {
    models: {
      "mac-qwen/qwen3.5": {
        alias: "qwen-mac"
      }
    }
  }
}

Then you can select it as:

qwen-mac

or the canonical ref:

mac-qwen/qwen3.5

4. Optional: make it a separate named agent

Add to agents.list:

1
2
3
4
5
6
7
8
9
{
  id: "qwen-mac",
  name: "Qwen Mac",
  model: "mac-qwen/qwen3.5",
  workspace: "/home/binia/.openclaw/workspace-main/",
  tools: {
    profile: "coding"
  }
}

Then OpenClaw can route/spawn work to that agent by agentId: "qwen-mac".


5. Restart/apply config

After editing:

openclaw gateway restart

Then verify:

openclaw models list | grep qwen

My take: use it first as a model alias (qwen-mac). Only add a separate agents.list entry if you want a different personality/tool policy/concurrency bucket. Otherwise you’re just adding config clutter.

Edit

Pub: 10 May 2026 06:57 UTC

Views: 2