Use OpenClaw

OpenClaw uses an OpenAI-compatible chat endpoint. Use the recommended sequence length, and raise either limit if OpenClaw’s prompt does not fit. Serve a model id without a slash, and enable automatic tool choice because OpenClaw sends tools.

edgellm serve Qwen/Qwen3.5-4B \
  --max-input-len 32768 \
  --max-kv-cache-capacity 40960 \
  --enable-auto-tool-choice \
  --served-model-name qwen35-local \
  --port 8000

Add this provider to ~/.openclaw/openclaw.json:

{
  models: {
    providers: {
      edgellm: {
        baseUrl: "http://127.0.0.1:8000/v1",
        api: "openai-completions",
        apiKey: "local",
        models: [
          {
            id: "qwen35-local",
            name: "Qwen3.5 4B",
            input: ["text"],
            contextWindow: 40960,
            maxTokens: 2048,
          },
        ],
      },
    },
  },
  agents: {
    defaults: {
      model: { primary: "edgellm/qwen35-local" },
    },
  },
}

baseUrl must end in /v1. The primary model id is edgellm/qwen35-local. That origin is trusted for this provider. Match apiKey to --api-key when the server requires one. The server ignores the bearer token when --api-key is omitted.