adolf/docs/providers/vllm.md

summary, read_when, title

summary	read_when	title
Run OpenClaw with vLLM (OpenAI-compatible local server)	You want to run OpenClaw against a local vLLM server You want OpenAI-compatible /v1 endpoints with your own models	vLLM

vLLM serves open-source (and some custom) models through an OpenAI-compatible HTTP API. OpenClaw connects using the openai-completions API and can auto-discover models when you opt in with VLLM_API_KEY.

Property	Value
Provider ID	vllm
API	openai-completions (OpenAI-compatible)
Auth	VLLM_API_KEY environment variable
Default base URL	http://127.0.0.1:8000/v1
Streaming usage	Supported (stream_options.include_usage)

Getting started

Your base URL must expose `/v1` endpoints (`/v1/models`, `/v1/chat/completions`). vLLM commonly runs on:

```text
http://127.0.0.1:8000/v1
```

Any non-empty value works if your server does not enforce auth:

```bash
export VLLM_API_KEY="vllm-local"
```

Replace with one of your vLLM model IDs:

```json5
{
  agents: {
    defaults: {
      model: { primary: "vllm/your-model-id" },
    },
  },
}
```

```bash openclaw models list --provider vllm ``` For non-interactive setup (CI, scripting), pass the base URL, key, and model directly:

openclaw onboard --non-interactive \
  --mode local \
  --auth-choice vllm \
  --custom-base-url "http://127.0.0.1:8000/v1" \
  --custom-api-key "vllm-local" \
  --custom-model-id "your-model-id"

Model discovery (implicit provider)

When VLLM_API_KEY is set (or an auth profile exists) and models.providers.vllm is not defined, OpenClaw queries GET http://127.0.0.1:8000/v1/models and converts the returned IDs into model entries.

If you set `models.providers.vllm` explicitly, OpenClaw uses only your declared models. Add `"vllm/*": {}` to `agents.defaults.models` to make OpenClaw also query that configured provider's `/models` endpoint and include all advertised vLLM models.

Explicit configuration

Configure explicitly when vLLM runs on a different host or port, you want to pin contextWindow/maxTokens, your server requires a real API key, or you connect to a trusted loopback, LAN, or Tailscale endpoint:

{
  models: {
    providers: {
      vllm: {
        baseUrl: "http://127.0.0.1:8000/v1",
        apiKey: "${VLLM_API_KEY}",
        api: "openai-completions",
        timeoutSeconds: 300, // Optional: extend request timeout for slow local models
        models: [
          {
            id: "your-model-id",
            name: "Local vLLM Model",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 8192,
          },
        ],
      },
    },
  },
}

To keep the provider dynamic without listing every model, add a wildcard to the visible model catalog:

{
  agents: {
    defaults: {
      models: {
        "vllm/*": {},
      },
    },
  },
}

Advanced configuration

vLLM is treated as a proxy-style OpenAI-compatible `/v1` backend, not a native OpenAI endpoint:

| Behavior                                | Applied?                         |
| --------------------------------------- | -------------------------------- |
| Native OpenAI request shaping           | No                               |
| `service_tier`                          | Not sent                         |
| Responses `store`                       | Not sent                         |
| Prompt-cache hints                      | Not sent                         |
| OpenAI reasoning-compat payload shaping | Not applied                      |
| Hidden OpenClaw attribution headers     | Not injected on custom base URLs |

For Qwen models, set `compat.thinkingFormat: "qwen-chat-template"` on the model row when the server expects Qwen chat-template kwargs. These models expose a binary `/think` profile (`off`, `on`) because Qwen chat-template thinking is an on/off flag, not an OpenAI-style effort ladder.

```json5
{
  models: {
    providers: {
      vllm: {
        models: [
          {
            id: "Qwen/Qwen3-8B",
            name: "Qwen3 8B",
            reasoning: true,
            compat: { thinkingFormat: "qwen-chat-template" },
          },
        ],
      },
    },
  },
}
```

OpenClaw maps `/think off` to:

```json
{
  "chat_template_kwargs": {
    "enable_thinking": false,
    "preserve_thinking": true
  }
}
```

Non-`off` thinking levels send `enable_thinking: true`. If your endpoint expects DashScope-style top-level flags instead, use `compat.thinkingFormat: "qwen"` to send `enable_thinking` at the request root.

For `vllm/nemotron-3-*` models with thinking off, the bundled plugin sends:

```json
{
  "chat_template_kwargs": {
    "enable_thinking": false,
    "force_nonempty_content": true
  }
}
```

To customize these values, set `chat_template_kwargs` under the model params. If you also set `params.extra_body.chat_template_kwargs`, that value wins because `extra_body` is the last request-body override.

```json5
{
  agents: {
    defaults: {
      models: {
        "vllm/nemotron-3-super": {
          params: {
            chat_template_kwargs: {
              enable_thinking: false,
              force_nonempty_content: true,
            },
          },
        },
      },
    },
  },
}
```

First confirm vLLM was started with the right tool-call parser and chat template for the model. vLLM documents `hermes` for Qwen2.5 models and `qwen3_xml` for Qwen3-Coder models.

Symptoms: skills/tools never run, the assistant prints raw JSON/XML such as `{"name":"read","arguments":...}`, or vLLM returns an empty `tool_calls` array when OpenClaw sends `tool_choice: "auto"`.

Some Qwen/vLLM combinations return structured tool calls only when the request uses `tool_choice: "required"`. Force it per model with `params.extra_body`:

```json5
{
  agents: {
    defaults: {
      models: {
        "vllm/Qwen-Qwen2.5-Coder-32B-Instruct": {
          params: {
            extra_body: {
              tool_choice: "required",
            },
          },
        },
      },
    },
  },
}
```

Replace the model id with the exact id from `openclaw models list --provider vllm`, or apply the same override from the CLI:

```bash
openclaw config set agents.defaults.models '{"vllm/Qwen-Qwen2.5-Coder-32B-Instruct":{"params":{"extra_body":{"tool_choice":"required"}}}}' --strict-json --merge
```

This is an opt-in workaround: it forces every turn with tools to make a tool call, so use it only for a dedicated model entry where that is acceptable. Do not set it as a global default for all vLLM models, and do not pair it with a proxy that converts arbitrary assistant text into executable tool calls.

If your vLLM server runs on a non-default host or port, set `baseUrl` in the explicit provider config:

```json5
{
  models: {
    providers: {
      vllm: {
        baseUrl: "http://192.168.1.50:9000/v1",
        apiKey: "${VLLM_API_KEY}",
        api: "openai-completions",
        timeoutSeconds: 300,
        models: [
          {
            id: "my-custom-model",
            name: "Remote vLLM Model",
            reasoning: false,
            input: ["text"],
            contextWindow: 64000,
            maxTokens: 4096,
          },
        ],
      },
    },
  },
}
```

Troubleshooting

For large local models, remote LAN hosts, or tailnet links, set a provider-scoped request timeout:

```json5
{
  models: {
    providers: {
      vllm: {
        baseUrl: "http://192.168.1.50:8000/v1",
        apiKey: "${VLLM_API_KEY}",
        api: "openai-completions",
        timeoutSeconds: 300,
        models: [{ id: "your-model-id", name: "Local vLLM Model" }],
      },
    },
  },
}
```

`timeoutSeconds` applies to vLLM model HTTP requests only: connection setup, response headers, body streaming, and the total guarded-fetch abort. It also raises the LLM idle/stream watchdog ceiling above the implicit ~120s default for this provider. Prefer this over increasing `agents.defaults.timeoutSeconds`, which controls the whole agent run.

Check that the vLLM server is running and accessible:

```bash
curl http://127.0.0.1:8000/v1/models
```

If you see a connection error, verify the host, port, and that vLLM started in OpenAI-compatible server mode. OpenClaw trusts the exact configured `models.providers.vllm.baseUrl` origin for guarded model requests on loopback, LAN, and Tailscale endpoints. Metadata/link-local origins remain blocked without explicit opt-in. Set `models.providers.vllm.request.allowPrivateNetwork: true` only when vLLM requests must reach another private origin, or `false` to opt out of exact-origin trust.

If requests fail with auth errors, set a real `VLLM_API_KEY` that matches your server configuration, or configure the provider explicitly under `models.providers.vllm`.

<Tip>
If your vLLM server does not enforce auth, any non-empty value for `VLLM_API_KEY` works as an opt-in signal for OpenClaw.
</Tip>

Auto-discovery requires `VLLM_API_KEY` to be set. If you have defined `models.providers.vllm`, OpenClaw uses only your declared models unless `agents.defaults.models` includes `"vllm/*": {}`. If a Qwen model prints JSON/XML tool syntax instead of executing a skill:

- Start vLLM with the correct parser/template for that model.
- Confirm the exact model id with `openclaw models list --provider vllm`.
- Add a dedicated per-model `params.extra_body.tool_choice: "required"` override only if `tool_choice: "auto"` still returns empty or text-only tool calls.

More help: [Troubleshooting](/help/troubleshooting) and [FAQ](/help/faq).

Related

Choosing providers, model refs, and failover behavior. Native OpenAI provider and OpenAI-compatible route behavior. Auth details and credential reuse rules. Common issues and how to resolve them.