Local CLI & owner-edge

The NOME CLI is a single binary named nome. On macOS and Linux the installer fetches the binary for your OS and CPU, verifies its checksum, and installs it to ~/.nome/bin/nome:

curl -fsSL https://nomadmaraud.com/cli/install.sh | bash

Add ~/.nome/bin to your PATH if the installer prompts you. Availability note: the public CLI download is gated per release channel and may show as coming soon until a channel is published for your platform — the installer fails safe rather than fetching an unverified binary.

Run nome login to sign in, then nome pair to make this machine a trusted runner (full flow in Pair a computer). Confirm with nome pair status. Run nome doctor to probe the host: it reports which CLIs are installed (codex, claude, git, docker, …) and their status, and it never reads or prints credentials.

A paired machine accepts work through the NOME host agent. On your Mac, start it from NOME by selecting This Mac as the Code host, or run the agent process directly so the machine stays available as a runner. The agent supervises each run, streams a normalized event envelope, and writes receipts.

The host agent never slurps dotfiles or scrapes environment variables for credentials — provider auth stays in each provider CLI's own config.

Pick This Mac in the Code host selector. Work runs against your local files and toolchain. If you signed in to a provider CLI on this Mac (for example the OpenAI Codex CLI), NOME can detect it and run a turn through that local seat — see Provider seats. Otherwise route through a local model, BYOK, or NOME credits.

If you use the OpenAI Codex CLI, sign in to it on this machine with its own flow (codex login). NOME detects the install and sign-in status through the CLI's own codex login status command — exit codes only, never credential material — and offers it as a route. You never open the Codex CLI directly; NOME launches it on this machine when you pick that seat. Use provider seats on your own trusted computer or private workspace.

A Linux machine you own — desktop, home server, or a box on your LAN — can be an owner-edge host. Install the CLI with the same one-liner, run nome login and nome pair, and select it as the Code host from NOME.

For a box that lives behind NAT or that you reach remotely, put it on a private network with Tailscale or reach it over SSH; NOME talks to the host agent over that reachable path. Keep the OS and CLI patched — an owner-edge host runs your code with your toolchain.

To keep an owner-edge host available across reboots, run the host agent under systemd as a user or system service. A unit that launches the agent with Restart=on-failure keeps it resilient. Inspect it with systemctl status and journalctl -u <unit> for logs.

The agent also has sleep-resistant session helpers so long runs survive transient interruptions; combined with a restart policy this gives you an always-on owner-edge runner.

Use nome pair status and nome doctor for a quick health read. Under systemd, journalctl has the agent's logs. The agent emits a normalized JSONL event stream per session; receipts and run state also surface in NOME, where the host shows online/offline. If a host shows offline in NOME, start the agent and confirm network reachability (and Tailscale/SSH for remote boxes).

Honest status: the CLI publishes a Windows binary (nome-win-x64.exe) and nome pair reports the windows platform, so pairing a Windows machine is on the supported path. Full host-agent parity on Windows (service install, owner-edge ergonomics) is still being hardened and is partially deferred — prefer macOS or Linux for an always-on host today, and treat Windows as best-effort until its host-agent lane is marked complete.