> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/relay/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/relay/_mcp/server.

# About

> Choose a local or remote NeMo Relay daemon deployment.

Use daemon mode to give coding agents a shared entry point for model requests
and hooks. A **coding harness** is the application that runs the agent, such as
Codex, Claude Code, or Pi. A **daemon** is a background process that stays running
between sessions. Each user-machine identity gets its own worker and runtime state.

## Start Here When

Use this section to:

* Run Relay at login or system startup.
* Share one local daemon across users on a computer.
* Connect client computers to a daemon on a Linux server.
* Distribute the same administrator-owned settings to every user.
* Operate workers, diagnose failed connections, and preserve streamed responses.

For a personal integration managed by `nemo-relay install`, use
[Coding Agent Installation](/nemo-relay-cli/plugin-installation). Managed daemon
mode uses different lifecycle commands and administrator-owned configuration.

## Choose a Deployment

| Deployment                     | Daemon runs as            | Available when                                                | Guide                                                              |
| ------------------------------ | ------------------------- | ------------------------------------------------------------- | ------------------------------------------------------------------ |
| Windows user task              | Signed-in user            | User is logged in                                             | [Windows](/daemon/windows)                                         |
| Windows system task or service | System service account    | Computer is running                                           | [Windows](/daemon/windows)                                         |
| macOS LaunchAgent              | Signed-in user            | User is logged in                                             | [macOS](/daemon/macos)                                             |
| macOS LaunchDaemon             | Dedicated service account | Computer is running                                           | [macOS](/daemon/macos)                                             |
| Linux user service             | User                      | User manager is running; optional lingering starts it at boot | [Linux](/daemon/linux)                                             |
| Linux system service           | Dedicated service account | Computer is running                                           | [Linux](/daemon/linux)                                             |
| Remote Linux daemon            | Server service account    | Server and private network are available                      | [Server](/daemon/remote-server), [clients](/daemon/remote-clients) |

Choose one service method per daemon endpoint. The default local endpoint is
`http://127.0.0.1:47632`. A personal Relay sidecar can already occupy that port.
Stop the existing integration before switching, or assign a different fixed port
and generate a separate managed bundle for that endpoint. Multiple user services
on one computer also need separate ports; one system service can serve all users.

## Before You Begin

Install a Relay binary that includes `nemo-relay daemon managed-bundle` on the
server and every client. Follow [Installation](/getting-started/installation)
for supported binaries and [Support Matrix](/reference/support-matrix) for
harness versions. Check `nemo-relay daemon --help` on each target before setup.

An administrator must provision the system configuration, stable executable,
and managed bundle, even for a user-level daemon. Workers ignore personal Relay
configuration. Users need a distinct route credential for each machine-user
identity, plus their normal model-provider authentication.

Remote deployments need a trusted HTTPS certificate and a private IPv4 route
in both directions: clients contact the daemon, and the daemon contacts client
workers. A connection that allows only outbound traffic from clients is not enough.

## Reading Order

1. Read [Architecture and Startup](/daemon/architecture).
2. Prepare [Configuration and Managed Clients](/daemon/configuration).
3. Follow the deployment guide for your operating system.
4. Complete [Verify Worker-Backed Operation](/daemon/operations#verify-worker-backed-operation).
5. Keep [Reference](/daemon/reference) available for command and lifecycle details.