openclaw/docs/nvidia.md

summary, read_when, title

summary	read_when	title
NVIDIA API setup for AI model access	You want to use NVIDIA's AI models You need NVIDIA_API_KEY setup	NVIDIA API

NVIDIA API

OpenClaw can use NVIDIA's API (https://integrate.api.nvidia.com/v1) for accessing various AI models. NVIDIA provides access to state-of-the-art language models through their integration endpoint.

API Setup

NVIDIA (direct)

• Base URL: https://integrate.api.nvidia.com/v1
• Environment variable: NVIDIA_API_KEY
• Get your API key from: NVIDIA NGC

Config example

{
  models: {
    providers: {
      nvidia: {
        apiKey: "nvapi-...",
        baseUrl: "https://integrate.api.nvidia.com/v1",
      },
    },
  },
  agents: {
    default: {
      provider: "nvidia",
      model: "nvidia/llama-3.1-nemotron-70b-instruct",
    },
  },
}

Available Models

OpenClaw includes support for several NVIDIA models:

• nvidia/llama-3.1-nemotron-70b-instruct (default) — High-performance instruction-following model
• nvidia/llama-3.3-70b-instruct — Latest Llama 3.3 variant
• nvidia/mistral-nemo-minitron-8b-8k-instruct — Smaller, efficient model

Environment Variable Setup

Set your NVIDIA API key as an environment variable:

export NVIDIA_API_KEY="nvapi-your-key-here"

Or add it to your .env file:

NVIDIA_API_KEY=nvapi-your-key-here

Usage in Config

Minimal configuration (uses environment variable):

{
  agents: {
    default: {
      provider: "nvidia",
      model: "nvidia/llama-3.1-nemotron-70b-instruct",
    },
  },
}

Explicit API key configuration:

{
  models: {
    providers: {
      nvidia: {
        apiKey: "NVIDIA_API_KEY",
        baseUrl: "https://integrate.api.nvidia.com/v1",
        api: "openai-completions",
      },
    },
  },
}

Professional and Personal Use

NVIDIA's API is suitable for both professional and personal applications:

• Professional: Enterprise-grade models for business applications, research, and development
• Personal: Access to powerful AI models for learning, experimentation, and personal projects

Notes

• NVIDIA API uses OpenAI-compatible endpoints
• Models are automatically discovered if NVIDIA_API_KEY is set
• Default context window: 131,072 tokens
• Default max tokens: 4,096 tokens