Skip to main content
This guide shows you how to configure Nvidia NIM as an external model provider in AI Studio. After setting up this provider, you can use foundation models from Nvidia’s API Catalog or your own self-hosted NIM deployments when building agents.
Nvidia NIM supports both cloud-hosted models via the Nvidia API Catalog and self-hosted deployments via Nvidia Inference Microservices. See Deployment options for details on each approach.

​
Prerequisites

Before adding Nvidia NIM models to AI Studio, you need:
AI Studio supports text generation and embedding models from Nvidia NIM. Video, audio, and image generation models are not supported.

​
Generate an NGC API key

Generate an API key from the NGC console to authenticate with Nvidia NIM.
  1. Navigate to the NGC API Key page
  2. Select Generate API Key
  3. Enter a description for your key (for example, AI Studio integration)
  4. Under Key Permissions, ensure NGC Catalog access is included in the Services Included list
  5. Select Generate
  6. Copy the API key immediately
The API key is only shown once. Copy it immediately and store it securely. If you lose the key, you must generate a new one.

​
Deployment options

Nvidia NIM offers two deployment approaches. Choose the one that matches your infrastructure requirements.

​
Cloud-hosted (Nvidia API Catalog)

Use Nvidia’s cloud-hosted models through the API Catalog for quick setup. This approach:
  • Requires no infrastructure management
  • Uses Nvidia’s hosted API endpoint (https://integrate.api.nvidia.com/v1)
  • Provides access to Nvidia-hosted and partner models available through the API Catalog
  • Bills usage through your Nvidia account

​
Self-hosted (Nvidia Inference Microservices)

Deploy NIM containers in your own infrastructure for greater control. This approach:
  • Runs models on your own GPU or cloud instances
  • Requires a custom API Base URL pointing to your NIM deployment
  • Provides greater control over where inference runs and how infrastructure is managed
  • Supports air-gapped or private network deployments
For self-hosted deployments, AI Studio must be able to send outbound HTTPS requests to your NIM endpoint. You may need to configure network rules or use a public endpoint.

​
Add Nvidia NIM models in AI Studio

After generating your API key, add Nvidia NIM models to AI Studio.
For self-hosted deployments, configure your NIM instance and note its API Base URL before proceeding. See Configure self-hosted endpoints for details.
  1. Navigate to Models & Guardrails > Models in AI Studio
  2. Select + Add model
  3. Select Nvidia NIM as the provider
  4. Enter your NGC API key
  5. Configure the endpoint:
    • For cloud-hosted models: Leave the API Base URL empty to use the default Nvidia endpoint
    • For self-hosted models: Enter your custom API Base URL (for example, https://your-nim-instance.example.com/v1)
  6. Select the models you want to add from the available list
  7. Configure team access:
    • All teams: Anyone with builder access can use the model
    • Specific teams: Restrict to selected teams
  8. Select Add model

​
Configure self-hosted endpoints

When using self-hosted NIM deployments, provide the base URL where your NIM instance is running. For detailed setup instructions, including hardware requirements and deployment options, see Get started with Nvidia NIM for LLMs.

​
API Base URL format

The API Base URL should point to your NIM deployment’s API endpoint: 
https://your-nim-host:port/v1
Replace your-nim-host and port with your actual NIM instance address.

​
Verify your endpoint

Before adding to AI Studio, verify your NIM endpoint is accessible: 
curl https://your-nim-instance.example.com/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"
A successful response returns a list of available models on your NIM instance. For more information on available endpoints, see the Nvidia NIM API reference.

​
Monitor costs

For cloud-hosted models, Nvidia bills usage directly to your Nvidia account based on the tokens processed. AI Studio also tracks usage and costs for external models, providing visibility into spending across all your models in one place. For self-hosted deployments, costs depend on your infrastructure (compute instances, resources). For information about monitoring model health and automatic recovery, see Monitor model health.

​
Troubleshoot Nvidia NIM configuration

​
Invalid API key error

If you see an “Invalid API key” or “Authentication failed” error:
  • Verify the API key is copied correctly without extra spaces
  • Check that the API key is still active in the NGC console
  • Ensure you’re using an NGC API key, not a different Nvidia credential type

​
Model not available error

If a model doesn’t appear or returns an error:
  • For cloud-hosted: Confirm the model is available in the Nvidia API Catalog
  • For self-hosted: Verify your NIM instance has the model deployed
  • Check that your API key has access to the specific model

​
Connection failed error

If AI Studio cannot connect to your NIM endpoint:
  • Verify the API Base URL is correct and includes the protocol (https://)
  • Check that your NIM instance is running and accessible
  • Ensure network rules allow traffic from AI Studio’s infrastructure
  • For self-hosted deployments, verify SSL certificates are valid

​
Unhealthy model status

If a model shows as unhealthy in AI Studio:
  • AI Studio automatically retries unhealthy models after a cooldown period
  • For cloud-hosted models experiencing transient issues, no action is needed
  • For self-hosted models, check that your NIM instance is running and healthy
  • For persistent issues, check the troubleshooting items above

​
Next steps