OpenClaw supports running with local Large Language Models (LLMs), allowing for:

• Privacy: Data never leaves your machine.
• Offline Use: Run without internet.
• Cost: No API fees.

LM Studio is the easiest way to serve OpenAI-compatible local models.

1. Install LM Studio Download from lmstudio.ai.

2. Load a Model Search for and download a model.

   • Recommendation: MiniMax-Text-01 or Llama-3-70B (if hardware permits).
   • Avoid small (<7B) quantized models for complex agent tasks.

3. Start Server In LM Studio, go to the Local Server tab and click Start Server.

   • Ensure it’s running on http://127.0.0.1:1234.

4. Configure OpenClaw Add the local provider to ~/.openclaw/openclaw.json.

   {
     models: {
       providers: {
         lmstudio: {
           baseUrl: "http://127.0.0.1:1234/v1",
           apiKey: "lm-studio", // Value doesn't matter
           models: [
             {
               id: "local-model", // Match the ID in LM Studio
               name: "My Local Model",
               contextWindow: 32000
             }
           ]
         }
       }
     },
     agents: {
       defaults: {
         model: { primary: "lmstudio/local-model" }
       }
     }
   }

You can use a local model for “easy” tasks (chat) and a hosted model (Claude/GPT-4) for complex reasoning or fallbacks.

{
  agents: {
    defaults: {
      model: {
        primary: "anthropic/claude-3-5-sonnet", // Smart cloud model
        fallbacks: ["lmstudio/local-model"]      // Local fallback
      }
    }
  }
}