NVIDIA Nemotron 3 Nano - Simple Usage Guide#

A comprehensive guide to using NVIDIA Nemotron 3 Nano for basic inference, reasoning modes, and building intelligent agents with LangChain.

Overview#

This notebook provides a streamlined walkthrough of NVIDIA Nemotron 3 Nano capabilities, from simple API calls to orchestrating multi-agent research systems.

Models Used#

LLM: nvidia/nemotron-3-nano-30b-a3b (via OpenRouter)

Key Features#

🔌 OpenAI-Compatible API via OpenRouter for easy integration
🧠 Reasoning Modes - Toggle chain-of-thought thinking ON/OFF
🔍 Web Search Agent with DuckDuckGo integration
💬 Conversation Memory using LangGraph’s InMemorySaver
🤝 Multi-Agent System with specialized agents (Search, Report Writer, Quality Reviewer)

What’s Covered#

Part	Topic	Description
1	Basic Usage	Simple API calls, streaming responses
2	Reasoning Modes	Enable/disable chain-of-thought reasoning
3	LangChain Agent	Research assistant with web search and memory
4	Multi-Agent System	Coordinator orchestrating specialist agents

Multi-Agent Architecture#

┌─────────────────┐
│   Coordinator   │ ◄── User Query
│   (Supervisor)  │
└────────┬────────┘
         │
    ┌────┴────┬──────────┐
    ▼         ▼          ▼
┌───────┐ ┌────────┐ ┌──────────┐
│Search │ │Report  │ │Quality   │
│Agent  │ │Writer  │ │Reviewer  │
└───────┘ └────────┘ └──────────┘

Requirements#

Python 3.13+
uv package manager
OpenRouter API key (get one here)

Quick Start#

Clone the repository
Install dependencies:
```
uv sync
```
Run the notebook and enter your OpenRouter API key when prompted

Resources#