### Install a single example package

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/contributing/index.html

Install a single example package using `uv pip install -e ./examples/<example_name>`. Example shown installs the Simple Calculator example.

```bash
uv pip install -e ./examples/getting_started/simple_web_query
```

--------------------------------

### Install nat_simple_web_query workflow

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/get-started/quick-start.html

Installs the simple web query workflow example using uv.

```bash
uv pip install -e examples/getting_started/simple_web_query
```

--------------------------------

### Install simple_calculator_observability workflow

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/observe/observe.html

Installs the simple_calculator example workflow used in the Weave guide.

```bash
uv pip install -e examples/observability/simple_calculator_observability
```

--------------------------------

### Install the simple calculator example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/improve-workflows/sizing-calc.html

Install the simple calculator example for the sizing calculator.

```bash
uv pip install -e examples/evaluation_and_profiling/simple_calculator_eval
```

--------------------------------

### Run simple calculator observability example with Arize AX

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/observe/observe.html

Installs the example package and runs the workflow with the Arize AX config. Expect a log line 'Started exporter 'arize_ax'' and traces in the Arize project.

```bash
uv pip install -e examples/observability/simple_calculator_observability/

nat run --config_file examples/observability/simple_calculator_observability/configs/config-arize-ax.yml --input "What is 2 * 4?"
```

--------------------------------

### Install NeMo Agent Toolkit and simple web query example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/build-workflows/llms/using-local-llms.html

Run from the repository root to install the toolkit and the simple web query example package.

```bash
uv pip install -e .
uv pip install -e examples/getting_started/simple_web_query
```

--------------------------------

### Install and Serve Simple Calculator Workflow

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/rest-api/api-server-endpoints.html

Installs the simple calculator example and starts the NeMo Agent Toolkit server on http://localhost:8000. Run these commands before using the API examples.

```bash
uv pip install -e examples/getting_started/simple_calculator
```

```bash
nat serve --config_file examples/getting_started/simple_calculator/configs/config.yml
```

--------------------------------

### Install nvidia-nat with uv

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/index.html

Install the NeMo Agent Toolkit package using the uv package manager.

```bash
uv pip install nvidia-nat
```

--------------------------------

### Install and Run simple_web_query Workflow with OTel+Dynatrace

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/observe/observe.html

Commands to install the simple_web_query example workflow and run it with OTel+Dynatrace telemetry settings. Telemetry data will appear in Dynatrace as the workflow runs.

```bash
# Install the workflow and plugins
uv pip install -e examples/getting_started/simple_web_query

# Run the workflow with OTel+Dynatrace telemetry settings
nat run --config_file examples/getting_started/simple_web_query/configs/config.yml --input "What is LangSmith?" 
```

--------------------------------

### Install from source with vLLM backend

Source: https://docs.nvidia.com/nemo/export-deploy/latest/index.html

Installs the package with the vLLM backend after installing prerequisites.

```bash
pip install --no-build-isolation .[vllm]
```

--------------------------------

### Install pre-commit hooks

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/contributing/index.html

Install and configure pre-commit hooks. First run takes longer than normal.

```bash
pre-commit install
```

--------------------------------

### nat start fastapi example command

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/cli.html

Launches a FastAPI server for a workflow using the specified configuration file, binding to host 0.0.0.0 on port 8000. After starting, Swagger API documentation is available at http://localhost:8000/docs.

```bash
nat start fastapi --config_file=path/to/config --host 0.0.0.0 --port 8000
```

--------------------------------

### Install and Run simple_calculator_observability Workflow with Galileo

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/observe/observe.html

Commands to install the simple_calculator_observability example and run it with Galileo telemetry settings. Configuration may need updating based on your Galileo account.

```bash
# Install the workflow and plugins
uv pip install -e examples/observability/simple_calculator_observability/

# Run the workflow with Galileo telemetry settings
# Note, you may have to update configuration settings based on your Galileo account
nat run --config_file examples/observability/simple_calculator_observability/configs/config-galileo.yml --input "What is 1*2?"
```

--------------------------------

### Set OpenAI API key environment variable

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/get-started/quick-start.html

Optional setup for workflows that require OpenAI. Replace <YOUR_OPENAI_API_KEY> with your actual key.

```bash
export OPENAI_API_KEY="<YOUR_OPENAI_API_KEY>"
```

--------------------------------

### Install from source with TRT-ONNX backend

Source: https://docs.nvidia.com/nemo/export-deploy/latest/index.html

Installs the package with the TRT-ONNX backend after installing prerequisites.

```bash
pip install --no-build-isolation .[trt-onnx]
```

--------------------------------

### Start server with --host and --port flags

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/components/auth/mcp-auth/index.html

Start the server with host and port matching the redirect_uri. The first command is for a remote server example; the second is for production behind a reverse proxy.

```bash
# For the remote server example above
nat serve --host 192.168.1.100 --port 8080

# Or for production with a reverse proxy
nat serve --host 0.0.0.0 --port 8000
```

--------------------------------

### Set NVIDIA API key environment variable

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/get-started/quick-start.html

Required before running any NeMo Agent Toolkit examples. Replace <YOUR_API_KEY> with your actual NVIDIA API key.

```bash
export NVIDIA_API_KEY=<YOUR_API_KEY>
```

--------------------------------

### Example nat optimize command with placeholder config path

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/improve-workflows/optimizer.html

Example invocation of the optimizer command. Replace the placeholder with the actual path to your configuration file. The command starts the optimization process and logs progress, parameters tested, and scores for each trial.

```bash
nat optimize --config_file <path to configuraiton file>
```

--------------------------------

### Install and launch NeMo AutoModel with uv pip and automodel CLI

Source: https://docs.nvidia.com/nemo/automodel/latest

Install the nemo-automodel package and run a training job using the automodel CLI with a YAML recipe. The example uses two processes per node and the llama3_2_1b_squad.yaml recipe.

```bash
uv pip install nemo-automodel
automodel --nproc-per-node=2 llama3_2_1b_squad.yaml
```

--------------------------------

### setup()

Source: https://docs.nvidia.com/nemo/rl/latest/index.html

Sets up the GRPO training environment.

```APIDOC
## setup()

### Description
Sets up the GRPO training environment.

```

--------------------------------

### setup() (Distillation)

Source: https://docs.nvidia.com/nemo/rl/latest/index.html

Sets up the distillation training environment.

```APIDOC
## setup() (Distillation)

### Description
Sets up the distillation training environment.

```

--------------------------------

### Install nvidia-nat-eval[full] for full eval support

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the full extra for nvidia-nat-eval to get complete eval functionality.

```bash
pip install "nvidia-nat-eval[full]"
```

--------------------------------

### Run nat_simple_web_query workflow

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/get-started/quick-start.html

Executes the simple web query workflow with a sample input.

```bash
nat run --config_file=examples/getting_started/simple_web_query/configs/config.yml --input "What is LangSmith"
```

--------------------------------

### Run Hello World example with nat CLI

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/index.html

Runs the Hello World example using the nat CLI and the workflow.yml file. The command outputs the workflow result to the console.

```bash
nat run --config_file workflow.yml --input "List five subspecies of Aardvarks"
```

--------------------------------

### Tavily Internet Search Configuration (Breaking)

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Configuration for the old Tavily internet search function. This is a breaking change in v1.8.0; workflows using this configuration no longer get Tavily search by installing only nvidia-nat[langchain].

```yaml
functions:
  internet_search:
    _type: tavily_internet_search
```

--------------------------------

### Basic finetuning command

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/improve-workflows/finetuning/rl_with_openpipe.html

Starts a finetuning run using the OpenPipe ART plugin. Requires a running ART server, an LLM with logprobs enabled, a training dataset, and a reward function.

```bash
nat finetune --config_file=configs/finetune.yml
```

--------------------------------

### Install nvidia-nat-security for nat red-team

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the package providing the `nat red-team` command.

```bash
pip install nvidia-nat-security
```

--------------------------------

### Access configuration in create_mcp_server()

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/extend/custom-components/mcp-server.html

Example showing how to access configuration via self.front_end_config (name, host, port, debug) and customize server creation based on debug mode.

```python
async def create_mcp_server(self) -> FastMCP:
    # Access server name from config
    server_name = self.front_end_config.name

    # Customize based on debug mode
    if self.front_end_config.debug:
        logger.info(f"Creating debug server: {server_name}")

    return FastMCP(
        name=server_name,
        host=self.front_end_config.host,
        port=self.front_end_config.port,
        debug=self.front_end_config.debug,
    )
```

--------------------------------

### Install nvidia-nat-eval CLI package

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the package providing the `nat eval` command.

```bash
pip install "nvidia-nat[eval]"
```

--------------------------------

### Install QEMU for multi-architecture CI

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/contributing/testing/running-ci-locally.html

Commands to install QEMU on an apt-based system for emulating a different architecture (e.g., arm64 on amd64).

```bash
sudo apt install qemu-utils qemu-system-arm qemu-user-static
```

--------------------------------

### Install nvidia-nat-ragas for RAGAS evaluator

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the RAGAS evaluator package when using `_type: ragas`.

```bash
pip install nvidia-nat-ragas
```

--------------------------------

### Curriculum Learning Training Schedule

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/improve-workflows/finetuning/concepts.html

Shows the training schedule for curriculum learning, where the model starts with the easiest 30% of examples and gradually expands to all examples over epochs.

```text
Epoch 1-5:   Train on easiest 30% of examples
Epoch 6-10:  Train on easiest 50% of examples
Epoch 11-15: Train on easiest 70% of examples
Epoch 16+:   Train on all examples
```

--------------------------------

### Hello World example output

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/index.html

Example console output from running the nat CLI command, showing the workflow result listing five subspecies of Aardvarks.

```text
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']
```

--------------------------------

### Install profiling support with nvidia-nat[profiler]

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the profiling package when needed. Use the extra or combine with eval.

```bash
pip install "nvidia-nat[profiler]"
```

```bash
pip install "nvidia-nat[eval, profiler]"
```

--------------------------------

### Create workflow.yml configuration file

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/index.html

Defines the agents, tools, and workflows for the Hello World example. Includes a wikipedia_search tool, a nim_llm LLM, and a react_agent workflow with verbose output and retry settings.

```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
```

--------------------------------

### Install nvidia-nat-config-optimizer package

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the optional package that owns the `nat optimize` command and optimizer implementation modules.

```bash
pip install nvidia-nat-config-optimizer
```

--------------------------------

### Install nvidia-nat[config-optimizer] extra

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install config optimizer support via the renamed extra (previously `nvidia-nat[optimizer]`).

```bash
pip install "nvidia-nat[config-optimizer]"
```

--------------------------------

### Example nat finetune commands with options

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/improve-workflows/finetuning/concepts.html

Example commands showing basic finetuning, overriding config values with -o, adding validation with --validation_dataset and --validation_interval, and using a remote endpoint with --endpoint and --endpoint_timeout. The --endpoint_timeout default is 300 seconds; --validation_interval default is 5 epochs.

```bash
# Basic finetuning
nat finetune --config_file=configs/finetune.yml

# Override number of epochs
nat finetune --config_file=configs/finetune.yml -o finetuning.num_epochs 20

# With validation
nat finetune --config_file=configs/finetune.yml \
    --validation_dataset=data/val.json \
    --validation_interval=3

# Using remote endpoint
nat finetune --config_file=configs/finetune.yml \
    --endpoint=http://localhost:8000/generate \
    --endpoint_timeout=600
```

--------------------------------

### A365 MCP Tooling Configuration Example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/api/nat/plugins/a365/tooling/tooling_config/index.html

YAML configuration for discovering and registering MCP servers from Agent 365. Requires the nvidia-nat-mcp package (install with `uv pip install nvidia-nat-mcp` or `uv pip install 'nvidia-nat[mcp]'`, or from source with `uv pip install 'nvidia-nat-a365[mcp]'`).

```yaml
function_groups:
  - type: a365_mcp_tooling
    agentic_app_id: "your-agent-id"
    auth_token: "your-auth-token"
```

--------------------------------

### Install from source with InFramework backend

Source: https://docs.nvidia.com/nemo/export-deploy/latest/index.html

Clones the Export-Deploy repository and installs the package with the InFramework backend.

```bash
git clone https://github.com/NVIDIA-NeMo/Export-Deploy
cd Export-Deploy/
pip install --no-build-isolation .
```

--------------------------------

### Install and run simple_calculator_observability with Phoenix

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/run-workflows/observe/observe.html

Installs the workflow and plugins, then runs the workflow with Phoenix telemetry settings. Telemetry data will start appearing in Phoenix as the workflow runs.

```bash
# Install the workflow and plugins
uv pip install -e examples/observability/simple_calculator_observability/

# Run the workflow with Phoenix telemetry settings
nat run --config_file examples/observability/simple_calculator_observability/configs/config-phoenix.yml --input "What is 1*2?"
```

--------------------------------

### nat start fastapi --help output

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/cli.html

Displays the help for the `nat start fastapi` command, listing all options for serving a workflow as a FastAPI microservice. The `--config_file` option is required and accepts a JSON/YAML file. Additional options include `--override` for dot-notation config overrides, `--host`, `--port`, `--reload`, `--workers`, `--step_adaptor`, `--workflow`, `--endpoints`, `--use_gunicorn`, and `--runner_class`.

```bash
$ nat start fastapi --help
Usage: nat start fastapi [OPTIONS]

Options:
  --config_file FILE              A JSON/YAML file that sets the parameters
                                  for the workflow.  [required]
  --override <TEXT TEXT>...       Override config values using dot notation
                                  (e.g., --override llms.nim_llm.temperature
                                  0.7)
  --root_path TEXT                The root path for the API
  --host TEXT                     Host to bind the server to
  --port INTEGER                  Port to bind the server to
  --reload BOOLEAN                Enable auto-reload for development
  --workers INTEGER               Number of workers to run
  --step_adaptor STEPADAPTORCONFIG
  --workflow ENDPOINTBASE         Endpoint for the default workflow.
  --endpoints ENDPOINT            Additional endpoints to add to the FastAPI
                                  app which run functions within the NAT
                                  configuration. Each endpoint must have a
                                  unique path.
  --use_gunicorn BOOLEAN          Use Gunicorn to run the FastAPI app
  --runner_class TEXT             The NAT runner class to use when launching
                                  the FastAPI app from multiple processes.
                                  Each runner is responsible for loading and
                                  running the NAT workflow. Note: This is
                                  different from the worker class used by
                                  Gunicorn.
  --help                          Show this message and exit.

```

--------------------------------

### Complete interactive workflow example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/build-workflows/advanced/interactive-workflows.html

Full example of an async function that acquires the manager, creates a prompt, prompts for input, and processes the response with error handling.

```python
async def _inner(prompt: str) -> str:
    try:
        context = Context.get()
        user_input_manager = context.user_interaction_manager

        human_prompt_text = HumanPromptText(text="Hello, how are you today?", required=True, placeholder="default")

        response = await user_input_manager.prompt_user_input(human_prompt_text)

        assert (isinstance(response.content, HumanResponseText))

        return response.content.text

    except Exception as e:
        logger.error("An error occurred when getting interaction content: %s", e)

        raise
```

--------------------------------

### Install plugin with pip

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/extend/third-party-plugins.html

Alternative installation method using pip.

```bash
pip install nemo-agent-toolkit-tavily
```

--------------------------------

### nat start console --help output

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/cli.html

Displays the help for the `nat start console` command, which runs a workflow from a configuration file with inputs from the command line or a file. The `--config_file` option is required. Options include `--override` for dot-notation overrides, `--input` for a single input, `--input_file` for a plain text file (batch evaluation should use `nat eval`), `--user_id` (defaults to 'nat_run_user_id'), and `--conversation_id` for isolating memory-backed runs.

```bash
$ nat start console --help
Usage: nat start console [OPTIONS]

Options:
  --config_file FILE         A JSON/YAML file that sets the parameters for the
                             workflow.  [required]
  --override <TEXT TEXT>...  Override config values using dot notation (e.g.,
                             --override llms.nim_llm.temperature 0.7)
  --input TEXT               A single input to submit the the workflow.
  --input_file FILE          Path to a plain text file containing a single
                             input to submit to the workflow. For batch
                             evaluation of multiple inputs, use 'nat eval'
                             instead.
  --user_id TEXT             User ID to use for the workflow session. Defaults
                             to 'nat_run_user_id' for single-user CLI
                             execution.
  --conversation_id TEXT     Conversation ID to use for the workflow session.
                             Set this to isolate memory-backed CLI runs.
  --help                     Show this message and exit.

```

--------------------------------

### Install NeMo Agent Toolkit with all groups and most extras

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/get-started/installation.html

Installs the library with developer tools (`--all-groups`) and most dependencies for profiling and plugins (`--extra most`). Use this for a full source installation.

```bash
uv sync --all-groups --extra most
```

--------------------------------

### Training Quickstart with Pre-configured Recipe

Source: https://docs.nvidia.com/nemo/megatron-bridge/latest/

Launch random-init pretraining using a pre-configured Llama 3.2 1B recipe on a single H100 GPU. Override cfg.model to use a different Hugging Face model ID or local directory, and optionally set cfg.checkpoint.pretrained_checkpoint for SFT/PEFT workflows.

```python
from megatron.bridge import AutoBridge
from megatron.bridge.recipes.llama.h100 import llama32_1b_pretrain_1gpu_h100_bf16_config
from megatron.bridge.training.gpt_step import forward_step
from megatron.bridge.training.pretrain import pretrain

if __name__ == "__main__":
    # The recipe uses the Llama 3.2 1B architecture from Hugging Face.
    # This is random-init pretraining and does not require a converted Megatron checkpoint.
    cfg = llama32_1b_pretrain_1gpu_h100_bf16_config()

    # The recipe already sets cfg.model internally using this pattern.
    # Override cfg.model to choose a different Hugging Face model ID as the architecture source.
    # cfg.model = AutoBridge.from_hf_pretrained("meta-llama/Llama-3.2-1B").to_megatron_provider(load_weights=False)

    # Optional: use a local Hugging Face model/config directory instead.
    # cfg.model = AutoBridge.from_hf_pretrained("/path/to/local/hf_model").to_megatron_provider(load_weights=False)

    # Optional: initialize weights from a converted Megatron checkpoint for SFT/PEFT
    # or other pretrained-weight workflows.
    # cfg.checkpoint.pretrained_checkpoint = "/path/to/megatron/checkpoint"

    # Override training parameters
    cfg.train.train_iters = 10
    cfg.scheduler.lr_decay_iters = 10000
    cfg.model.vocab_size = 8192
    cfg.tokenizer.vocab_size = cfg.model.vocab_size

    pretrain(cfg, forward_step)
```

--------------------------------

### ToolTestRunner usage example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/api/nat/test/tool_test_runner/index.html

Shows how to instantiate ToolTestRunner and test a tool with minimal setup or with mocked dependencies using the with_mocks() context manager. The example is from the class docstring and demonstrates the expected usage pattern.

```python
>>> runner = ToolTestRunner()
>>> # Test a tool with minimal setup
>>> result = await runner.test_tool(
...     config_type=MyToolConfig,
...     config_params={"param1": "value1"},
...     input_data="test input"
... )
>>> # Test a tool with mocked dependencies
>>> async with runner.with_mocks() as mock_builder:
>>>     mock_builder.mock_llm("my_llm", "mocked response")
>>>     result = await runner.test_tool(
...         config_type=MyToolConfig,
...         config_params={"llm_name": "my_llm"},
...         input_data="test input"
...     )
```

--------------------------------

### Start built container

Source: https://docs.nvidia.com/nemo/export-deploy/latest/index.html

Runs the container built with the Dockerfile, mounting the current directory and enabling GPU access.

```bash
docker run --rm -it -w /workdir -v $(pwd):/workdir \
  --entrypoint bash \
  --gpus all \
  nemo-export-deploy
```

--------------------------------

### Install nvidia-nat-profiler for performance evaluators

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/resources/migration-guide.html

Install the profiler package when using performance evaluators or profiling workflows. Note: the previous dependency group name `nvidia-nat-profiling` is now `nvidia-nat-profiler`.

```bash
pip install nvidia-nat-profiler
```

--------------------------------

### get_outstanding_step_count()

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/api/nat/builder/intermediate_step_manager/index.html

Get the number of outstanding (started but not ended) steps.

```APIDOC
## get_outstanding_step_count()

### Description
Get the number of outstanding (started but not ended) steps.

### Method
get_outstanding_step_count

### Returns
int: Number of steps that have been started but not yet ended
```

--------------------------------

### Show nat run help and options

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/cli.html

Displays the help output for the `nat run` command, which is an alias for `nat start console`. The help lists all available options including `--config_file`, `--override`, `--input`, and `--input_file`. The `--config_file` option is required.

```bash
$ nat run --help
Usage: nat run [OPTIONS]

Options:
  --config_file FILE         A JSON/YAML file that sets the parameters for the
                             workflow.  [required]
  --override <TEXT TEXT>...  Override config values using dot notation (e.g.,
                             --override llms.nim_llm.temperature 0.7)
  --input TEXT               A single input to submit the the workflow.
  --input_file FILE          Path to a json file of inputs to submit to the
                             workflow.
  --help                     Show this message and exit.
```

--------------------------------

### Reinstall workflow example

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/reference/cli.html

Example command to reinstall a workflow named 'my_workflow' after updating dependencies in pyproject.toml. The reinstall process rebuilds the package, uninstalls the existing version, installs the updated version, and verifies registered components.

```bash
nat workflow reinstall my_workflow
```

--------------------------------

### Install plugin with uv

Source: https://docs.nvidia.com/nemo/agent-toolkit/latest/extend/third-party-plugins.html

Install a third-party plugin package into the same Python environment as nvidia-nat-core using uv.

```bash
uv add nemo-agent-toolkit-tavily
```