NeMo Data Designer

Install via pip for the quickest start, or clone the repository for development and run 'make install'. The pip package name is data-designer.

# Via pip
pip install data-designer

# From source
git clone https://github.com/NVIDIA-NeMo/DataDesigner.git
cd DataDesigner && make install

Export the API key for your chosen provider. At least one of NVIDIA_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY is required. You can configure multiple providers and select between them at runtime.

export NVIDIA_API_KEY="your-nvidia-api-key"
export OPENAI_API_KEY="your-openai-api-key"
export OPENROUTER_API_KEY="your-openrouter-api-key"

Use the data-designer CLI to list configured providers and available models, confirming your API keys are recognized correctly.

data-designer config providers
data-designer config models
data-designer config list

Add the Data Designer skill to Claude Code to generate datasets using natural language directly in your coding sessions.

npx skills add NVIDIA-NeMo/DataDesigner

Add Data Designer to your MCP client configuration to expose data generation capabilities as tools accessible from Claude Desktop or other MCP clients.

{
  "mcpServers": {
    "datadesigner": {
      "command": "npx",
      "args": ["datadesigner"],
      "env": {
        "NVIDIA_API_KEY": "your-nvidia-api-key"
      }
    }
  }
}

Use the data-designer Python API or CLI to define your dataset schema and generate records. Use preview mode first to test your configuration before a full-scale run.

# Python API example
from data_designer import DataDesigner

dd = DataDesigner()
dd.add_field("question", description="A factual question about world history")
dd.add_field("answer", description="A concise accurate answer", depends_on=["question"])
result = dd.generate(num_records=100, preview=True)