NVIDIA NeMo Agent Toolkit

What is NVIDIA NeMo Agent Toolkit about?

The toolkit provides a unifying interface for building, composing, and managing AI agents across different frameworks (LangChain, LlamaIndex, CrewAI, Semantic Kernel, etc.). It focuses on reusability, rapid development, profiling, observability, evaluation, and full Model Context Protocol (MCP) support.

How to use NVIDIA NeMo Agent Toolkit?

Prerequisites: Git, Git LFS, uv (≥0.5.4), Python 3.11/3.12.

Clone the repository:

git clone git@github.com:NVIDIA/NeMo-Agent-Toolkit.git nemo-agent-toolkit
cd nemo-agent-toolkit

Initialize submodules:

git submodule update --init --recursive

Fetch LFS files:

git lfs install
git lfs fetch
git lfs pull

Create a virtual environment:

uv venv --seed .venv --python 3.11   # or 3.12
source .venv/bin/activate

Install the library (core only):

uv sync

To install all optional groups and plugins:

uv sync --all-groups --all-extras
uv pip install -e '.[<plugin_name>]'   # e.g., uv pip install -e '.[langchain]'

Verify installation:
```
nat --version
```
Run a workflow (example Hello World):
- Create workflow.yaml defining tools, LLM, and a react_agent.
- Export NVIDIA_API_KEY.
- Execute:
```
nat run --config_file workflow.yaml --input "List five subspecies of Aardvarks"
```

Key Features

Framework Agnostic – works alongside LangChain, LlamaIndex, CrewAI, Semantic Kernel, or custom Python agents.
Reusability – agents, tools, and workflows are function calls that can be composed and reused.
Rapid Development – start from pre‑built agents/tools and customize them.
Profiling – detailed token and timing metrics at workflow, tool, and agent levels.
Observability – integrations with Phoenix, Weave, Langfuse and OpenTelemetry‑compatible platforms.
Evaluation System – built‑in utilities to validate workflow accuracy.
User Interface – chat UI for interactive debugging and visualization.
Full MCP Support – can act as MCP client or server to publish/consume remote tools.

Use Cases

Building enterprise‑grade AI assistants that need to call internal APIs or databases.
Creating reusable AI‑agent components for multiple products within an organization.
Profiling and optimizing costly LLM calls in complex agentic pipelines.
Monitoring agent behavior in production with observability platforms.
Deploying custom tools via MCP to allow remote agents to access them securely.

FAQ

Q: Do I need to switch my existing agent framework? A: No. The toolkit sits on top of or alongside existing frameworks without requiring migration.

Q: Which Python versions are supported? A: Python 3.11 and 3.12.

Q: How do I add a new plugin (e.g., for LangChain)? A: Install it from the packages directory, e.g., uv pip install -e '.[langchain]'.

Q: Can I use the toolkit without any optional extras? A: Yes. Installing with uv sync provides the core functionality; extras add profiling, UI, and MCP capabilities.

Q: How do I run the toolkit as an MCP server? A: Follow the documentation under docs/source/workflows/mcp/mcp-server.md to expose your tools via MCP.

Q: Where can I find more examples? A: The examples folder in the repository contains ready‑to‑run workflows and detailed READMEs.

NVIDIA NeMo Agent toolkit is a flexible, lightweight, and unifying library that allows you to easily connect existing enterprise agents to data sources and tools across any framework.

[!NOTE] NeMo Agent toolkit was previously known as the Agent Intelligence (AIQ) toolkit, and AgentIQ. The library was renamed to better reflect the purpose of the toolkit and to align with the NVIDIA NeMo family of products. The core technologies, performance and roadmap remain unchanged and the API is fully compatible with previous releases.

The rename is still in progress and references to the previous name may still be found in the codebase and documentation.

Key Features

Framework Agnostic: NeMo Agent toolkit works side-by-side and around existing agentic frameworks, such as LangChain, LlamaIndex, CrewAI, and Microsoft Semantic Kernel, as well as customer enterprise frameworks and simple Python agents. This allows you to use your current technology stack without replatforming. NeMo Agent toolkit complements any existing agentic framework or memory tool you're using and isn't tied to any specific agentic framework, long-term memory, or data source.
Reusability: Every agent, tool, and agentic workflow in this library exists as a function call that works together in complex software applications. The composability between these agents, tools, and workflows allows you to build once and reuse in different scenarios.
Rapid Development: Start with a pre-built agent, tool, or workflow, and customize it to your needs. This allows you and your development teams to move quickly if you're already developing with agents.
Profiling: Use the profiler to profile entire workflows down to the tool and agent level, track input/output tokens and timings, and identify bottlenecks. While we encourage you to wrap (decorate) every tool and agent to get the most out of the profiler, you have the freedom to integrate your tools, agents, and workflows to whatever level you want. You start small and go to where you believe you'll see the most value and expand from there.
Observability: Monitor and debug your workflows with dedicated integrations for popular observability platforms such as Phoenix, Weave, and Langfuse, plus compatibility with OpenTelemetry-based observability platforms. Track performance, trace execution flows, and gain insights into your agent behaviors.
Evaluation System: Validate and maintain accuracy of agentic workflows with built-in evaluation tools.
User Interface: Use the NeMo Agent toolkit UI chat interface to interact with your agents, visualize output, and debug workflows.
Full MCP Support: Compatible with Model Context Protocol (MCP). You can use NeMo Agent toolkit as an MCP client to connect to and use tools served by remote MCP servers. You can also use NeMo Agent toolkit as an MCP server to publish tools via MCP.

With NeMo Agent toolkit, you can move quickly, experiment freely, and ensure reliability across all your agent-driven projects.

Component Overview

The following diagram illustrates the key components of NeMo Agent toolkit and how they interact. It provides a high-level view of the architecture, including agents, plugins, workflows, and user interfaces. Use this as a reference to understand how to integrate and extend NeMo Agent toolkit in your projects.

NeMo Agent toolkit Components Diagram

Get Started

Prerequisites

Before you begin using NeMo Agent toolkit, ensure that you meet the following software prerequisites.

Install Git
Install Git Large File Storage (LFS)
Install uv (version 0.5.4 or later, latest version is recommended)
Install Python (3.11 or 3.12)

Install From Source

Clone the NeMo Agent toolkit repository to your local machine.

git clone git@github.com:NVIDIA/NeMo-Agent-Toolkit.git nemo-agent-toolkit
cd nemo-agent-toolkit

Initialize, fetch, and update submodules in the Git repository.
```
git submodule update --init --recursive
```

Fetch the data sets by downloading the LFS files.

git lfs install
git lfs fetch
git lfs pull

Create a Python environment.
```
uv venv --seed .venv
source .venv/bin/activate
```
Make sure the environment is built with Python version 3.11 or 3.12. If you have multiple Python versions installed, you can specify the desired version using the --python flag. For example, to use Python 3.11:
```
uv venv --seed .venv --python 3.11
```
You can replace --python 3.11 with any other Python version (3.11 or 3.12) that you have installed.
Install the NeMo Agent toolkit library. To install the NeMo Agent toolkit library along with all of the optional dependencies. Including developer tools (--all-groups) and all of the dependencies needed for profiling and plugins (--all-extras) in the source repository, run the following:
```
uv sync --all-groups --all-extras
```
Alternatively to install just the core NeMo Agent toolkit without any plugins, run the following:
```
uv sync
```
At this point individual plugins, which are located under the packages directory, can be installed with the following command uv pip install -e '.[<plugin_name>]'. For example, to install the langchain plugin, run the following:
```
uv pip install -e '.[langchain]'
```
[!NOTE] Many of the example workflows require plugins, and following the documented steps in one of these examples will in turn install the necessary plugins. For example following the steps in the examples/getting_started/simple_web_query/README.md guide will install the nvidia-nat-langchain plugin if you haven't already done so.

In addition to plugins, there are optional dependencies needed for profiling. To install these dependencies, run the following:
```
uv pip install -e '.[profiling]'
```
Verify the installation using the NeMo Agent toolkit CLI
```
nat --version
```
This should output the NeMo Agent toolkit version which is currently installed.

Hello World Example

Ensure you have set the NVIDIA_API_KEY environment variable to allow the example to use NVIDIA NIMs. An API key can be obtained by visiting build.nvidia.com and creating an account.
```
export NVIDIA_API_KEY=<your_api_key>
```

Create the NeMo Agent toolkit workflow configuration file. This file will define the agents, tools, and workflows that will be used in the example. Save the following as workflow.yaml:

functions:
   # Add a tool to search wikipedia
   wikipedia_search:
      _type: wiki_search
      max_results: 2

llms:
   # Tell NeMo Agent toolkit which LLM to use for the agent
   nim_llm:
      _type: nim
      model_name: meta/llama-3.1-70b-instruct
      temperature: 0.0

workflow:
   # Use an agent that 'reasons' and 'acts'
   _type: react_agent
   # Give it access to our wikipedia search tool
   tool_names: [wikipedia_search]
   # Tell it which LLM to use
   llm_name: nim_llm
   # Make it verbose
   verbose: true
   # Retry up to 3 times
   parse_agent_response_max_retries: 3

Run the Hello World example using the nat CLI and the workflow.yaml file.

nat run --config_file workflow.yaml --input "List five subspecies of Aardvarks"

This will run the workflow and output the results to the console.

Workflow Result:
['Here are five subspecies of Aardvarks:\n\n1. Orycteropus afer afer (Southern aardvark)\n2. O. a. adametzi  Grote, 1921 (Western aardvark)\n3. O. a. aethiopicus  Sundevall, 1843\n4. O. a. angolensis  Zukowsky & Haltenorth, 1957\n5. O. a. erikssoni  Lönnberg, 1906']

Feedback

We would love to hear from you! Please file an issue on GitHub if you have any feedback or feature requests.

Acknowledgements

We would like to thank the following open source projects that made NeMo Agent toolkit possible:

NVIDIA NeMo Agent Toolkit