Install NVIDIA AI Workbench
Install AI Workbench on your DGX Spark system and complete the initial setup wizard.
On your DGX Spark, open the NVIDIA AI Workbench application and click "Begin Installation".
- The installation wizard will prompt for authentication
- Wait for the automated install to complete (several minutes)
- Click "Let's Get Started" when installation finishes
NOTE
If you encounter the following error message, reboot your DGX Spark and then reopen NVIDIA AI Workbench: "An error occurred ... container tool failed to reach ready state. try again: docker is not running"
Verify API key requirements
Next, you should ensure you have both required API keys before proceeding with the project setup. Keep these keys safe!
- Tavily API Key: https://tavily.com/
- NVIDIA API Key: https://org.ngc.nvidia.com/setup/api-keys
- Ensure this key has
Public API Endpointspermissions
Keep both keys available for the next step.
Clone the agentic RAG project
You'll then clone the pre-built agentic RAG project from GitHub into your AI Workbench environment.
From the AI Workbench landing page, select the Local location, if not done so already, then click "Clone Project" from the top right corner.
Paste this Git repository URL in the clone dialog: https://github.com/NVIDIA/workbench-example-agentic-rag
Click "Clone" to begin the clone and build process.
Configure project secrets
You can then configure the API keys required for the agentic RAG application to function properly.
While the project builds, configure the API keys using the yellow warning banner that appears:
- Click "Configure" in the yellow banner
- Enter your
NVIDIA_API_KEY - Enter your
TAVILY_API_KEY - Save the configuration
Wait for the project build to complete before proceeding.
Launch the chat application
You can now start the web-based chat interface where you can interact with the agentic RAG system.
Navigate to Environment > Project Container > Apps > Chat and start the web application.
A browser window will open automatically and load with the Gradio chat interface.
Test the basic functionality
Verify the agentic RAG system is working by submitting a sample query.
In the chat application, click on or type a sample query such as: How do I add an integration in the CLI?
Wait for the agentic system to process and respond. The response, while general, should demonstrate intelligent routing and evaluation.
Validate project
Confirm your setup is working correctly by testing the core features.
Verify the following components are functioning:
- Web application loads without errors
- Sample queries return responses
- No API authentication errors appear
- The agentic reasoning process is visible in the interface under "Monitor"
Complete optional quickstart
You can evaluate advanced features by uploading data, retrieving context, and testing custom queries.
Substep A: Upload sample dataset Complete the in-app quickstart instructions to upload the sample dataset and test improved RAG-based responses.
Substep B: Test custom dataset (optional) Upload a custom dataset, adjust the Router prompt, and submit custom queries to test customization.
Cleanup and rollback
You can remove the project if needed.
WARNING
This will permanently delete the project and all associated data.
To remove the project completely:
- In AI Workbench, click on the three dots next to a project
- Select "Delete Project"
- Confirm deletion when prompted
NOTE
All changes are contained within AI Workbench. No system-level modifications were made outside the AI Workbench environment.
Next steps
You can also explore further advanced features and development options with the agentic RAG system:
- Modify component prompts in the project code
- Upload different documents to test routing and customization
- Experiment with different query types and complexity levels
- Review the agentic reasoning logs in the "Monitor" tab to understand decision-making
Consider customizing the Gradio UI or integrating the agentic RAG components into your own projects.