Skip to content

docs: Add FAQ section for common questions about NeMo Agent Toolkit #1930

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
meichuanyi wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

docs: Add FAQ section for common questions about NeMo Agent Toolkit #1930

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
159 changes: 159 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,6 +167,165 @@ Before getting started, it's possible to run this simple workflow and many other
- [ ] MCP authentication improvements.
- [ ] Improved memory interface to support self-improving agents.

## ❓ Frequently Asked Questions (FAQ)

### What is NVIDIA NeMo Agent Toolkit?

NVIDIA NeMo Agent Toolkit (NAT) is an enterprise-grade toolkit that adds intelligence to AI agents across any framework—enhancing speed, accuracy, and decision-making through instrumentation, observability, and continuous learning. It works side-by-side with popular frameworks like LangChain, LlamaIndex, CrewAI, Microsoft Semantic Kernel, and Google ADK.
Copy link
Copy Markdown

@coderabbitai coderabbitai Bot May 14, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Replace NAT acronym usage in prose to satisfy docs lint rules.

NAT appears in narrative text in several places; this violates the Markdown vocabulary/style rules and may fail Vale. Use “NeMo Agent Toolkit” (or “the toolkit”), and reserve lowercase backticked nat only for technical identifiers/CLI references.

As per coding guidelines, “Documentation in Markdown files should not use NAT as an acronym… The exception to this rule is when referring to package names or code identifiers that contain "nat", these should always be surrounded by backticks…and be lowercase.”

Also applies to: 181-181, 257-257, 264-264, 287-287

🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@README.md` at line 174, Replace narrative uses of the acronym "NAT" with the
full name "NeMo Agent Toolkit" (or "the toolkit" where concise) and only use the
lowercase backticked form `nat` for package names or code/CLI identifiers;
update every prose occurrence of "NAT" in the README (including the sentence
currently using "NAT" in the description of the toolkit) to follow this rule and
keep capitalization and punctuation consistent.


### How is NeMo Agent Toolkit different from LangChain or CrewAI?

NeMo Agent Toolkit is **framework-agnostic** and works **alongside** these frameworks, not as a replacement:
- **LangChain/CrewAI** provide agent orchestration and workflow logic
- **NeMo Agent Toolkit** adds instrumentation, profiling, evaluation, optimization, and fine-tuning capabilities
- You can use NAT with LangChain, CrewAI, LlamaIndex, Semantic Kernel, Google ADK, or custom agents

### What LLM providers are supported?

NeMo Agent Toolkit supports:
- **NVIDIA NIMs** (nvidia/nemotron models, llama models, etc.)
- **OpenAI** (GPT-4, GPT-4o, etc.)
- **Anthropic** (Claude models)
- **Google Gemini**
- **Azure OpenAI**
- **AWS Bedrock**
- **Local models** via vLLM, Ollama, and other OpenAI-compatible endpoints

### How do I install NeMo Agent Toolkit?

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

For framework-specific integrations:
```bash
pip install "nvidia-nat[langchain]" # LangChain/LangGraph integration
pip install "nvidia-nat[crewai]" # CrewAI integration
pip install "nvidia-nat[llama_index]" # LlamaIndex integration
```

### What is the Profiler and how does it help?

The [Profiler](./docs/source/improve-workflows/profiler.md) provides:
- **Agent-level profiling** — trace entire agent execution flows
- **Token-level profiling** — analyze token efficiency and costs
- **Bottleneck identification** — find slow nodes, redundant calls, expensive operations
- **Performance optimization guidance** — actionable recommendations

### What is the Evaluation System?

The [Evaluation System](./docs/source/improve-workflows/evaluate.md) helps:
- **Validate agent accuracy** — test outputs against expected results
- **Maintain workflow quality** — regression testing for agent changes
- **Offline evaluation** — evaluate without production deployment
- **Multi-dimensional metrics** — accuracy, latency, cost, user satisfaction

### What is Hyper-Parameter and Prompt Optimizer?

The [Optimizer](./docs/source/improve-workflows/optimizer.md) automatically:
- **Tunes hyper-parameters** — temperature, max_tokens, top_p, etc.
- **Optimizes prompts** — finds the best prompt templates
- **Runs experiments** — systematic search across configuration space
- **Reports best configurations** — recommendations for optimal performance

### Can I fine-tune LLMs for my specific agent?

Yes! [Fine-tuning with Reinforcement Learning](./docs/source/improve-workflows/finetuning/index.md) allows:
- **Train intrinsic workflow knowledge** into the model
- **Improve agent-specific performance** — better reasoning for your domain
- **Reduce hallucinations** — more accurate outputs
- **Lower inference costs** — smaller models can perform better after fine-tuning

### What is Dynamo Runtime Intelligence?

[Dynamo Runtime Intelligence](./examples/dynamo_integration/latency_sensitivity_demo/README.md) provides:
- **Latency sensitivity inference** — automatically detect per-request latency requirements
- **Cache control hints** — intelligent KV-cache optimization
- **Load-aware routing** — route requests to optimal servers
- **Priority-aware serving** — prioritize latency-sensitive requests

### What is Agent Performance Primitives (APP)?

[Agent Performance Primitives (APP)](https://docs.langchain.com/oss/python/integrations/providers/nvidia#install-2) accelerate graph-based agent frameworks:
- **Parallel execution** — run independent nodes simultaneously
- **Speculative branching** — precompute likely future paths
- **Node-level priority routing** — prioritize critical nodes
- Works with **LangChain, CrewAI, Agno** and other graph-based frameworks

### Does NeMo Agent Toolkit support MCP?

Yes! NAT provides full [Model Context Protocol (MCP)](./docs/source/build-workflows/mcp-client.md) support:
- **MCP Client** — integrate MCP tools into your agents
- **MCP Server** — serve your tools and agents as MCP servers
- [FastMCP Workflow Publishing](./docs/source/run-workflows/fastmcp-server.md) — publish workflows as MCP servers

### Does NeMo Agent Toolkit support A2A Protocol?

Yes! NAT supports the [Agent-to-Agent (A2A) Protocol](./docs/source/components/integrations/a2a.md) for:
- **Distributed agent teams** — multiple agents working together
- **Authentication support** — secure inter-agent communication
- **Cross-framework collaboration** — agents from different frameworks can cooperate

### Is there a UI for interacting with agents?

Yes! The [Built-In User Interface](./docs/source/run-workflows/launching-ui.md) provides:
- **Chat interface** — interact with your agents
- **Visual output** — see execution results
- **Workflow debugging** — inspect intermediate steps
- **Real-time monitoring** — observe agent behavior

### How do I observe agent execution in production?

NeMo Agent Toolkit provides multiple [Observability](./docs/source/run-workflows/observe/observe.md) options:
- **Native LangSmith Integration** — trace execution, run experiments, compare outcomes
- **OpenTelemetry** — standard observability signals
- **Phoenix** — Arize AI observability platform
- **Custom exporters** — integrate with your monitoring stack

### What telemetry does NeMo Agent Toolkit collect?

See [Telemetry](#-telemetry) section for details. NAT collects:
- **Anonymous usage statistics** — feature adoption, workflow success rates
- **Performance metrics** — latency, token efficiency
- **Error classification** — failure categories
- **No sensitive data** — prompts, outputs, file paths, config values are never collected

### Can I disable telemetry?

Yes. Set the environment variable:
```bash
export NAT_TELEMETRY_ENABLED=false
```

### What Python versions are supported?

NeMo Agent Toolkit requires **Python 3.11, 3.12, or 3.13**.

### Where can I find examples?

See the [Examples](./examples/README.md) directory for:
- Hello World workflow
- LangChain integration examples
- CrewAI integration examples
- Dynamo integration demos
- MCP server/client examples
- Evaluation and optimization tutorials

### How do I get help?

- **Documentation**: [docs.nvidia.com/nemo/agent-toolkit](https://docs.nvidia.com/nemo/agent-toolkit/latest)
- **Troubleshooting**: [docs/resources/troubleshooting.md](./docs/source/resources/troubleshooting.md)
- **GitHub Issues**: [NVIDIA/NeMo-Agent-Toolkit/issues](https://github.com/NVIDIA/NeMo-Agent-Toolkit/issues)
- **Examples**: [examples/](./examples/README.md)

### How can I contribute?

See [Contributing Guide](./docs/source/resources/contributing/index.md) for:
- Development environment setup
- Code style guidelines
- PR submission process
- Testing requirements

## 📊 Telemetry

The NeMo Agent Toolkit includes runtime telemetry hooks for the `nat` command-line tool to help guide improvements. Telemetry is best-effort and never blocks or fails a CLI invocation. Once you opt in (see below), events are sent to the shared NeMo Usage Telemetry ingest.
Expand Down