> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/dsx-air/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/dsx-air/_mcp/server.

# Air SDK

This project provides a Python SDK for interacting with the NVIDIA Air API.

## Prerequisite

The SDK requires Python 3.10. There are a number of ways to achieve this such as `poetry`, `uv` or `virtualenv`.
The following shows how to achieve this on a few popular OS's using `virtualenv`.

```
# Install Python 3.10
brew install python@3.10  # MacOS
apt-get install python3.10  # Ubuntu/Debian
sudo pacman -S python310  # Arch

# Install, create and activate a Python 3.10 virtual environment
python3.10 -m pip install virtualenv
python3.10 -m virtualenv venv
. venv/bin/activate
```

## Installation (WIP)

1. Clone this repo to `/path/to/air-sdk`
2. Navigate to `/path/to/air-sdk`
3. `pip install -e .`

## Authentication Options (WIP)

1. NGC Device Login with `AirApi.with_device_login(email=<YOUR_EMAIL>, org_num=<YOUR_NGC_ORG_NUMBER>)`
2. NGC CLI Login with `AirApi.with_ngc_config()` (or just `AirApi()`)
3. Explicitly specify an NGC API Key (also known as a Starfleet API Key, or "SAK") by instantiating `AirApi.with_api_key(api_key=<YOUR_STARFLEET_API_KEY>)`

To use options 2 or 3 you will need to **not be on the Northeast GlobalProtect region**

And to use option 2 you will need to have the [NGC CLI](https://org.stg.ngc.nvidia.com/setup/installers/cli) installed and to have configured it with `ngc config set`. This will create a file at `~/.ngc/config` which contains your NGC credentials that will be auto-detected by the Air SDK.

We do not support NGC service keys yet.

## Examples

[Examples overview](/sdk/v1-0-0-examples)

## API Reference

[air\_sdk API reference](/sdk/v1-0-0-air-sdk)