Skip to main content

​
NVIDIA NIM

NVIDIA NIM provides optimised, containerised model inference through an OpenAI-compatible API endpoint. Models are tuned for NVIDIA GPU hardware and served via NVIDIA’s cloud infrastructure.

​
Prerequisites

Create an API key at build.nvidia.com and set it on the gateway host: 
echo 'NVIDIA_API_KEY=nvapi-...' >> ~/.openclaw/.env

​
Configuration

NVIDIA NIM uses a custom base URL that must be set explicitly: 
{
  models: {
    providers: {
      nvidia: {
        apiKey: "NVIDIA_API_KEY", // references the env var
        baseUrl: "https://integrate.api.nvidia.com/v1",
        models: [
          "meta/llama-3.3-70b-instruct",
          "nvidia/llama-3.1-nemotron-70b-instruct",
          "mistralai/mistral-large-2-instruct",
        ],
      },
    },
  },
}
apiKey is the name of the environment variable, not the key value itself. Keep secrets in .env, not in committed config files.

​
Selecting a model

{
  agents: {
    defaults: {
      model: { primary: "nvidia/meta/llama-3.3-70b-instruct" },
    },
  },
}
ModelNotes
meta/llama-3.3-70b-instructMeta Llama 3.3 70B — strong general-purpose
nvidia/llama-3.1-nemotron-70b-instructNVIDIA fine-tuned Nemotron variant
mistralai/mistral-large-2-instructMistral Large 2 on NVIDIA infrastructure
NVIDIA NIM model IDs use a vendor/model-name format. Use the full ID as shown in the table above.
Run openclaw models list --provider nvidia for the current catalog.

​
Troubleshooting

The API key is invalid. Generate a new key at build.nvidia.com and update ~/.openclaw/.env.
Confirm the full model ID including the vendor prefix (e.g. meta/llama-3.3-70b-instruct). NVIDIA NIM requires the scoped name.
Confirm baseUrl is set to https://integrate.api.nvidia.com/v1 in your provider config — the provider will not work without it.