Is this page helpful?

NIM Metadata API Prerequisites#

Review the following requirements before you make your first request to the NIM Metadata API. The read endpoints are open, so the requirements are limited to an HTTP client and the identifiers for the NIM microservice you want to inspect.

Hardware#

Reading metadata requires no GPU and no particular hardware. Any machine that can issue an HTTPS request can query the API, because the service returns recorded facts rather than running inference.

You do need details about the GPUs you intend to deploy on, because those values are what you compare the profiles against. The following table describes the hardware details to collect:

Detail

How to Obtain It

GPU PCI identifier

Run lspci -nn | grep -i nvidia on Linux.

vRAM per GPU

Run nvidia-smi --query-gpu=memory.total --format=csv.

Number of available GPUs

Count the devices you intend to dedicate to the deployment.

NVIDIA driver version

Run nvidia-smi --query-gpu=driver_version --format=csv.

For the procedure that uses these values, refer to Choose a Profile for Your Hardware.

Software#

The following table describes the software you need:

Requirement

Details

HTTP client

Any HTTP client, such as curl, Postman, or the HTTP library for your language. The examples in this documentation use curl and Python.

JSON parsing

Responses use the application/json media type. Parse permissively so that unrecognized keys do not cause a failure.

Network Access#

Requests go to the following base URL:

https://api.ngc.nvidia.com/v2

All examples in this documentation use this host.

Credentials#

You do not need credentials. The endpoints that use the HTTP GET method are open, so you do not need any of the following:

  • An NGC API key

  • An NGC account

  • An NVIDIA AI Enterprise license or any other subscription

Metadata is readable for any NIM microservice whose metadata has been published, whether or not you are entitled to download that NIM microservice. This is deliberate, because the purpose of the API is to let you evaluate a NIM microservice before you commit to it.

Note

Access to the metadata is not the same as access to the NIM microservice. Being able to read the profiles for a NIM microservice does not grant you the right to pull the container. Pulling still follows the normal entitlement rules for that container.

NIM Microservice Identifiers#

Every request needs two things: which container, and which tag. You can find both on the NGC Catalog page for the NIM microservice.

Resource Identifier#

A resourceId identifies the container. It is a slash-separated path in one of the following forms:

  • {org}/{repo-name}, which has two segments

  • {org}/{team-name}/{repo-name}, which has three segments

Both forms are valid. The form that applies depends on whether the repository for the NIM microservice sits directly under the organization or inside a team within it. The service fixes the organization segment, and for public NIM microservices it is nim.

The following table describes both forms:

NIM Microservice Repository

Resource Identifier

openfold3 under the organization root

nim/openfold3

llama-3.1-8b-instruct under the team meta

nim/meta/llama-3.1-8b-instruct

Tag#

The tag is the container tag, exactly as it appears in the catalog, such as 1.2.0, 2.0.12, or latest.

Tip

If you can pull the container as nvcr.io/nim/meta/llama-3.1-8b-instruct:2.0.12, then your resourceId is nim/meta/llama-3.1-8b-instruct and your tag is 2.0.12.

Metadata Coverage#

Metadata is attached to a NIM microservice when it is published, so coverage grows as NIM microservices are released and re-released. Not every NIM microservice has metadata today. NIM microservices published before this API launched in September 2026 generally do not have API-backed metadata, and will not until they are republished.

A NIM microservice without metadata returns a 404 status code. That is a normal, expected response rather than an error in your request. For guidance on distinguishing it from other 404 responses, refer to Troubleshooting. For guidance on handling it in an integration, refer to Notes for Automated Agents.

Pre-Request Checklist#

Confirm the following before you make your first request:

  • HTTP client: You have an HTTP client that can issue a request that uses the GET method.

  • Network access: You can reach https://api.ngc.nvidia.com/v2 from your network.

  • Resource identifier: You have the resourceId for the NIM microservice, with either two or three segments and no colon in the value.

  • Tag: You have the container tag as a separate value, not appended to the resourceId.

  • Status code handling: You accept a 404 status code as a normal outcome that indicates no published metadata.

Next Steps#

Now that you have confirmed the requirements, explore the following resources: