NVIDIA NeMo Relay
NVIDIA/NeMo-Relay
Agent runtime and library for scopes, lifecycle events and middleware on tool and LLM calls
NVIDIA NeMo Relay is a multi-language agent runtime and library that gives visibility into and control over agent runs without changing the existing agent stack. Hermes Agent understands its plugin configurations and has Relay built in.
What NVIDIA NeMo Relay does
NeMo Relay provides a shared runtime for scopes, policy, plugins and lifecycle events that coding agents, applications, framework integrations, middleware and observability backends can use. The README says it adds visibility and control without requiring changes to the existing agent stack. Its CLI can run Codex or Claude Code through Relay and write raw events and normalized trajectory files to disk.
Observability exporters cover ATOF events, ATIF trajectories and OpenTelemetry traces, logs and metrics. Supported integrations include LangChain, LangGraph, Deep Agents and OpenClaw, and reusable middleware or exporters can be packaged as plugins. Packages are published on PyPI, npm and crates.io. For Hermes Agent, the README says NeMo Relay is built in, so no separate observability plugin or Relay CLI setup is required.
Key features
- Shared runtime for scopes, policy, plugins and lifecycle events across agents, apps and frameworks
- CLI that runs Codex or Claude Code through Relay and writes raw events and trajectory files
- Exports ATOF events, ATIF trajectories and OpenTelemetry traces, logs and metrics
- Integrations for LangChain, LangGraph, Deep Agents and OpenClaw
- Plugin system for packaging reusable middleware or exporters
When to use it
- Recording agent trajectories to disk to inspect and debug a run
- Sending agent traces, logs and metrics to an OpenTelemetry backend
- Adding middleware around tool and LLM calls in a framework integration
Who it is for: Developers and platform teams who want observability and policy control over coding agents and agent applications.
How it fits with Hermes Agent
Hermes Agent understands NeMo Relay plugin configurations, and the README says Relay is built into Hermes Agent, so no separate setup is needed there.
How to install NVIDIA NeMo Relay
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
pip install nemo-relay-cli-bin
curl -fsSL https://raw.githubusercontent.com/NVIDIA/NeMo-Relay/main/install.sh | sh
nemo-relay --versionRequirements: The installer supports Linux x86_64 and ARM64, macOS Apple Silicon, and Windows x86_64 and ARM64.
FAQ
What is NVIDIA NeMo Relay?
NVIDIA NeMo Relay is a multi-language agent runtime and library for execution scope management, lifecycle events and middleware on tool and LLM calls. It is written in Rust.
Does NVIDIA NeMo Relay work with Hermes Agent?
Yes, Hermes Agent understands NeMo Relay plugin configurations. The README says Relay is built into Hermes Agent, so no separate observability plugin or Relay CLI setup is required.
How do I install NVIDIA NeMo Relay?
Install the prebuilt CLI with pip install nemo-relay-cli-bin, or run the install.sh script, then check it with nemo-relay --version. A PowerShell installer is available for Windows.
Similar dev tools for Hermes Agent
All dev toolsSDK for building voice and text devices for Hermes Agent, with ESP32 firmware, a Linux client and a simulator
letta-ai trajectoryTypeScript library that normalizes agent session transcripts from Hermes and other runtimes into one format
VintLin Skill FlowCLI and macOS app for installing, updating and sharing agent skills across many coding agents
missuo TokensCLI and web leaderboard that tracks AI coding-assistant token usage across 41 clients
VAMFI Agentic SubstratePortable agent setup from one INSTALL.md with minimal instructions and reversible changes
Tencent SkillHoneRepairs failing agent skills from real runs and leaves a tested local PR for review
Related guides: How to install Hermes Agent · Run multiple Hermes agents with profiles