nvidia-ai-api-cli

Command-line chat client for the NVIDIA NIM / Integrate API. Works on Linux and Windows.

Stream model replies in the terminal, switch models with /model, and remember the last selected model between sessions.

• Repo: github.com/evrenater/nvidia-ai-api-cli
• Default model: nvidia/nemotron-3-ultra-550b-a55b
• API endpoint: https://integrate.api.nvidia.com/v1/chat/completions
• License: GPL-3.0

---

Features

• Streaming chat against NVIDIA-hosted models
• Live model catalog from GET /v1/models
• /model picker (list, filter, select by number or id)
• Bare number after a model list selects that entry
• Optional thinking / reasoning stream
• Syntax-highlighted fenced code blocks in the terminal
• I didn't want to store model information in user data location, better use terminal flag.
• Optional Rich UI (falls back to ANSI if missing)

---

Requirements

• Python 3.9+
• Free NVIDIA API key from build.nvidia.com

Dependencies

Package	Required	Purpose
requests	Yes	HTTP client for the NVIDIA API
rich	Optional	Panel banner and nicer prompts

pip install -r requirements.txt

Or:

pip install requests rich prompt_toolkit

---

Install

git clone https://github.com/evrenater/nvidia-ai-api-cli.git
cd nvidia-ai-api-cli

Virtual environment (recommended)

Linux / macOS (Not tested on macOS)

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Windows (PowerShell)

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Windows (cmd)

python -m venv .venv
.venv\Scripts\activate.bat
pip install -r requirements.txt

---

API key

Create a key at build.nvidia.com.

Linux / macOS (current shell)

export NVIDIA_API_KEY="nvapi-..."

Persist it in ~/.bashrc or ~/.zshrc if you want it every session.

Windows PowerShell (current session)

$env:NVIDIA_API_KEY = "nvapi-..."

Windows PowerShell (permanent for your user)

[System.Environment]::SetEnvironmentVariable("NVIDIA_API_KEY", "nvapi-...", "User")

Open a new terminal after setting a permanent variable.

Do not commit your key or put it in source files.

---

Run

Linux / macOS

python3 chat.py

Windows

python chat.py

Optional flags (beta)

python chat.py --model nvidia/nemotron-3-super-120b-a12b
python chat.py --thinking
python chat.py --hide-thinking
python chat.py --no-stream
python chat.py --max-tokens 2048
python chat.py --temperature 0.7
python chat.py --system "You are a concise coding assistant."

Flag	Default	Description
--model	saved / default	Model id
--max-tokens	4096	Max completion tokens
--temperature	1.0	Sampling temperature
--top-p	0.95	Nucleus sampling
--timeout	300	Request timeout (seconds)
--thinking	off	Request thinking when the model allows
--hide-thinking	off	Hide reasoning stream
--no-stream	off	Non-streaming replies
--system	none	System prompt for this session

---

In-chat commands

Command	Action
/help	Show help
/model	List chat models
/model <n>	Select by number from the last list
/model <id>	Select by full model id
/model <filter>	Filter list (example: /model llama)
/quit /exit /bye /q	Exit (Ctrl+C also works)

After /model prints a numbered list, typing 64 and Enter selects model #64 (only while a pick is pending).

---

Project layout

nvidia-ai-api-cli/
├── chat.py              # CLI entry point
├── requirements.txt
├── nvidia-ai-api-cli.png
├── LICENSE              # GPL-3.0
└── README.md

---

Notes

• GET /v1/models returns the full NVIDIA catalog (chat, embeddings, vision, ASR, and more). The CLI hides non-chat entries in the default list; a full id still works with /model <id>.
• Not every catalog model supports chat/completions. If a selection fails, try another model.
• Keep NVIDIA_API_KEY out of git. Add a local .env only if you load it yourself; do not commit secrets.

---

License

This project is licensed under the GNU General Public License v3.0.