nemo-relay-get-started

Guides first-time NeMo Relay users through CLI, integration, or language quick-start paths.

3.2k|370|Updated Feb 25, 2026

Search Tags:#CLI #plugins #Onboarding #Quick Start #agent instrumentation #Observability #nemo relay

Ready to run

npx skills add https://github.com/NVIDIA/skills --skill nemo-relay-get-started

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: nemo-relay-get-started
Source: https://github.com/NVIDIA/skills/tree/main/skills/nemo-relay-get-started
Command: npx skills add https://github.com/NVIDIA/skills --skill nemo-relay-get-started

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

New NeMo Relay users often struggle to pick the right starting point and risk jumping into production configuration before seeing any value. This Skill selects the least-complex supported quick-start path and verifies one observable success before any production setup.

Core Features & Use Cases

Ordered Try-Now Path Selection: Chooses among the CLI transparent-run path, the Hermes Agent native path, built-in integrations (LangChain, LangGraph, Deep Agents, OpenClaw), or manual Python, Node.js, and Rust instrumentation based on the user's environment.
First-Value Verification: Configures local ATOF and ATIF observability output, runs a safe read-only smoke trial, and confirms root, tool, and LLM lifecycle events as proof Relay is active.
Guided Plugin Progression: After the first proof, recommends exactly one next plugin (Adaptive, NeMo Guardrails, PII Redaction, or Model Pricing) and hands off to install, instrumentation, or observability skills when needed.
Use Case: A developer with a LangGraph application asks to try Relay; the Skill routes to the maintained LangGraph integration reference instead of manual instrumentation and verifies captured lifecycle events.

Quick Start

Ask your agent to help you try NeMo Relay for the first time and it will pick the simplest quick-start path for your setup.

Frequently Asked Questions about nemo-relay-get-started

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I get started with NeMo Relay for the first time?▼

Start with the CLI try-now path, which wraps Codex or Claude Code in a transparent Relay session without changing application code. Verify nemo-relay --version, configure local observability output, and run one read-only smoke prompt to see captured events.

How do I try NeMo Relay with a LangChain or LangGraph application?▼

Use the built-in integrations path, which attaches Relay through the framework's maintained middleware or callbacks rather than manual instrumentation. Install the matching integration package and follow the maintained LangChain or LangGraph integration guide for the smallest attachment point.

Does Hermes Agent require a separate NeMo Relay installation?▼

No, NeMo Relay is built into Hermes Agent and requires no separate Relay installation or observability plugin setup. Hermes Agent understands NeMo Relay plugin configurations natively, so no additional quick-start steps apply.

Why are my Relay ATOF or ATIF output files empty after a trial run?▼

Empty output usually means a plugin activation, output-directory permission, hook loading, or session finalization issue rather than a broken install. Run nemo-relay doctor <agent> --json and check supported user configuration, writable output directories, and whether the agent session fully ended.

Can I use repository-local .nemo-relay config files for Relay configuration?▼

No, repository-local .nemo-relay/config.toml and plugins.toml are unsupported as active Relay configuration. Relay uses the XDG user configuration directory, and local .nemo-relay/atof and .nemo-relay/atif paths are output artifacts, not configuration layers.

When should I not use the NeMo Relay quick start?▼

Do not use the quick start for production-wide rollout with OTLP backends, guardrails, model pricing, or adaptive tuning in one pass. Prove one local attachment boundary first, then hand off production concerns to the dedicated instrumentation and plugin configuration workflows.