nemo-relay-install — Claude Code skill: install Relay
The nemo-relay-install skill identifies your installation target and deploys the current stable NeMo Relay package, resolving prerequisites and verifying availability. Works across agent runtimes, frameworks, and language bindings.
Apache-2.0 227 installs ★ 4,106 updated 2026-10-23
npx skills add NVIDIA/skills --skill nemo-relay-install -a claude-codeWhat nemo-relay-install does
nemo-relay-install asks a single clarifying question to determine whether you need Relay for an existing agent runtime, your own application, or the CLI. It then selects the matching installation path and resolves the latest stable release from metadata rather than stored version pins.
The skill reads installation docs for your chosen target, derives requirements from the release's support matrix and package metadata, and verifies the environment before proceeding. It shows the install command, handles environment setup, and confirms availability—without configuring export, launching a gateway, or making model calls as a completion check.
For Codex Desktop with persistent storage, the skill flags any risk of provider changes and requests confirmation before global modifications. Once installation succeeds, it hands off to nemo-relay-get-started with your target, release, and verified environment ready.
What it helps with
- choose the right Relay installation type for your agent or framework
- find current stable release requirements before installing
- install Relay and verify it works in your environment
- avoid version mismatches with docs and third-party bundles
- resolve dependencies for your platform
Example prompts for the agent
- Set up NeMo Relay for Claude Code in my project
- I have an agent framework on Python—how do I add Relay to it?
- What's the latest stable Relay and what do I need to install it?
How to install nemo-relay-install
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill nemo-relay-install -a claude-code2Manually: put the files in place
~/.claude/skills/nemo-relay-install/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/nemo-relay-install/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill nemo-relay-install -a codex2Manually: put the files in place
.agents/skills/nemo-relay-install/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill nemo-relay-install -a cursor2Manually: put the files in place
.cursor/skills/nemo-relay-install/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install NVIDIA/skills skills/nemo-relay-install2Manually: put the files in place
.github/skills/nemo-relay-install/SKILL.md1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill nemo-relay-install -a gemini-cli2Platform CLI
gemini skills install https://github.com/NVIDIA/skills --path skills/nemo-relay-installManually: put the files in place
.devin/skills/nemo-relay-install/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill nemo-relay-install -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about nemo-relay-install
How do I install nemo-relay-install in Claude Code?
Run npx skills add NVIDIA/skills --skill nemo-relay-install -a claude-code, then check ~/.claude/skills/nemo-relay-install/SKILL.md for details.
What does nemo-relay-install do?
The skill identifies where you need Relay—agent runtime, application binding, CLI, or local gateway—selects the right installation steps, resolves the current stable release and its requirements, then installs and verifies.
Why does the skill ask me what I'm installing Relay into?
The installation path differs for agent integration, language bindings, and CLI. A Python project does not automatically mean you want a Python binding if you're adding Relay to a coding agent.
Source text
SKILL.md · Apache-2.0
--- name: nemo-relay-install description: Use this skill when choosing or running NeMo Relay installation for an agent runtime, application language, framework integration, or plugin bundle, or checking a runtime's native Relay readiness. Do not use it for first-run instrumentation or export configuration. license: Apache-2.0 metadata: author: NVIDIA Corporation and Affiliates --- # Install NeMo Relay Resolve what the user needs, install the appropriate current stable package or executable, verify availability, and stop. Installation does not include a first application, plugin configuration, scopes, or a model-consuming trial. ## Disambiguate Before Installing Establish the desired outcome and the boundary that owns execution before selecting a Relay type. Inspect available project context, but do not assume that a Python manifest means the user wants a Python binding rather than Relay for their coding agent. When the target is unclear, ask one concise question: > Are you adding Relay to an existing agent runtime or framework, integrating > it into your own application, or using its CLI or local gateway? Resolve the named runtime, framework, language, or plugin only as far as needed to select its installation instructions. Preserve an explicit choice or an earlier handoff; do not repeat questions already answered. A generic "install Relay" or "try Relay" request alone does not authorize choosing the CLI. | Established target | Reference | |---|---| | CLI, coding-agent wrapper, or persistent host integration | [CLI Installation](references/cli-install.md) | | Native agent integration, framework integration, or plugin bundle | [Maintained Integrations](references/maintained-integrations.md) | | Application-owned language binding | [Language Packages](references/language-packages.md) | | Explicit contributor or source-build request | Current repository development guide | ## Resolve Current Stable Requirements Read the current [installation guide](https://docs.nvidia.com/nemo/relay/getting-started/installation) for the selected target. Resolve prerequisites from the selected release's installation sections, support matrix, and package metadata rather than a remembered standalone prerequisites page. Confirm the latest published stable release using [release metadata](https://api.github.com/repos/NVIDIA/NeMo-Relay/releases/latest), [official releases](https://github.com/NVIDIA/NeMo-Relay/releases/latest), and the relevant package registry. Check the target's actual platform assets and package availability, not just a repository version or a pending PR. - Default to the latest published stable release, excluding prereleases unless explicitly requested. Preserve explicit user or project version constraints. - Derive operating-system, architecture, language, host, and dependency requirements from that release's documentation and metadata at execution time. - Do not copy version pins or minimum-version tables into this skill. Retain a specific version note only for an established incompatibility, with its source. - A quick-start example's pin does not override the stable default. If its requirements conflict with the selected release or project, explain the mismatch and resolve the choice before installing. - Native or dependency-managed integrations may own their Relay dependency. Follow that supported relationship rather than installing a second runtime. - Resolve independently released bundles separately and check compatibility. Their release number need not match Relay's. Resolve the documentation version explicitly: an unversioned page may display an older stable guide. When rendered documentation is inaccessible or lacks the selected path, use the official repository at the matching release tag; use its development branch only for an explicitly requested development workflow. A cached release redirect alone does not establish the latest version. If current sources cannot be checked, state what is unverified. Do not infer the newest release or requirements from this skill, an old checkout, a generated benchmark, or a source-built quick-start test. ## Install And Verify Inspect the environment, existing installation, manifest, and authoritative package manager. Load only the relevant reference and current target-specific documentation. Preserve the project's manager and virtual environment. Derive the install command from those sources, preview it, and proceed within existing authorization. For remote installers, separate download, review, and execution; use the resolved release's installer and integrity checks rather than executing an unreviewed default-branch response. Use only the documented availability check for the chosen target. Diagnose a failed check at that boundary instead of reinstalling unrelated packages. Report the resolved release, installation location, and verification evidence. Do not configure export, launch a gateway, or make a model call as an install check. ## Persistent Codex Desktop Continuity Persistent Codex setup can change the provider used to display Desktop conversations. Before that path, read the continuity section in [CLI Installation](references/cli-install.md), check the referenced upstream issue's applicability, and explain any relevant risk. Retain the recovery-note and explicit confirmation gate before global changes. Do not inspect, copy, delete, or rewrite private session storage or SQLite state as a visibility workaround. Use supported Relay install, uninstall, and diagnostic commands. ## Hand Off For a first working example, hand off to `nemo-relay-get-started` with the chosen target, resolved release, environment, and verified installation. Continue from the appropriate guide step without repeating selection. For durable export, instrumentation expansion, or plugin behavior, use the matching workflow. Do not repeat installation when it is already complete.
© NVIDIA Corporation and Affiliates · Apache-2.0 · NVIDIA/skills