How to Use DeepSeek-V4-Pro for Free Using OpenClaude & NVIDIA NIM

DeepSeek-V4-Pro is a powerhouse model, and integrating it into your local OpenClaude workspace can supercharge your AI workflows. Thanks to NVIDIA NIM (NVIDIA Inference Microservices), you can access an OpenAI-compatible endpoint for DeepSeek models—often utilising NVIDIA’s free developer credits or open-access tiers—directly within OpenClaude.

Here is the exact setup process to get deepseek-ai/deepseek-v4-pro running in your environment.

Prerequisites

• OpenClaude installed and running on your machine.
• An NVIDIA Developer account to generate your API key.

Step 1: Generate Your NVIDIA API Key

To route OpenClaude through NVIDIA’s servers, you need an API key from their integration platform.

1. Head over to the NVIDIA API Catalog (or the integrate.api.nvidia.com dashboard).
2. Find the DeepSeek-V4-Pro model in the catalogue.
3. Click on Get API Key (or generate one from your account settings).
4. Copy this key (it usually starts with nvapi- or is a long alphanumeric string like the one in your terminal). Keep it secure.

Step 2: Configure Your OpenClaude Settings

OpenClaude allows you to define custom model providers by editing your configuration files (typically settings.json or your environment config). You will need to map the NVIDIA NIM endpoint and supply your API key.

Open your settings.json file and locate the "customModelProviders" section. Add the NVIDIA NIM configuration like this, or you can ask your AI to do it for you as well:

JSON

{
  "customModelProviders": {
    "nvidia-nim": {
      "baseUrl": "https://integrate.api.nvidia.com/v1",
      "apiKey": "YOUR_NVIDIA_API_KEY",
      "models": {
        "deepseek-v4-pro": "deepseek-ai/deepseek-v4-pro"
      }
    }
  }
}

Note: Make sure to replace "YOUR_NVIDIA_API_KEY" with the actual key you generated in Step 1.

Step 3: Configure Agent Routing (If using Agent Models)

If you are utilizing OpenClaude’s agent capabilities and want to set DeepSeek-V4-Pro as the default brain for your automated tasks, you need to update the "agentModels" and "agentRouting" configuration.

Add or update the following block in your configuration:

JSON

{
  "agentModels": {
    "nvidia-deepseek": {
      "base_url": "https://integrate.api.nvidia.com/v1",
      "api_key": "YOUR_NVIDIA_API_KEY"
    }
  },
  "agentRouting": {
    "default": "nvidia-deepseek"
  }
}

What This Does:

• base_url: Points OpenClaude away from default Anthropic/OpenAI servers and directly to NVIDIA’s high-performance API endpoint.
• agentRouting: Ensures that any default agent prompt is automatically routed through the DeepSeek model hosted on NVIDIA NIM.

Step 4: Verify the Connection

Once your JSON files are saved, restart OpenClaude. You can verify the connection is working by initiating a chat or checking the terminal output. Under the hood, OpenClaude will now format its requests to match NVIDIA’s endpoint requirements, allowing you to use DeepSeek-V4-Pro with full configuration controls (like temperature, top_p, and high max_tokens limits).

You are now fully set up to leverage DeepSeek’s advanced reasoning capabilities within your custom OpenClaude environment!