First run

After installing, enroll the device into your Waired network with waired init. Run it once per machine.

Same account = same network

Sign in to every device with the same Google account. Devices enrolled under the same account join the same private network and can reach each other; that shared identity is what lets one machine use another’s model.

Enroll the device

Linux / macOS

sudo waired init

Windows

& "C:\Program Files\Waired\waired.exe" init

waired init walks you through three prompts:

1. Google sign-in. A browser opens to complete the sign-in. (Add --no-browser to print the URL and code instead of opening a browser — useful on a headless server.)

2. Enable the inference engine? (Y/n) — Choose Y on a machine that should run a model (your GPU box). The default is Y when the machine has enough memory (roughly 8 GB of VRAM) and N otherwise. On a client laptop that only uses a remote model, choose N.

3. Share with mesh peers? (Y/n) — Choose Y on a server so your other devices can send it inference requests. Choose N to keep the engine private to this machine (you can still use it yourself). This is independent of the engine toggle — see sharing vs. pausing.

If you enabled the engine, init looks at this machine’s CPU, memory, and GPU and automatically picks the best bundled coding model that fits — a larger, higher-quality model on a capable box, a smaller one on a modest laptop. It checks free disk first (and steps down to a smaller model, or skips the download, rather than filling the disk), then pulls the chosen model and waits for it in the foreground, showing download progress. The pull can take a few minutes; once it finishes, init runs a quick benchmark and — if this machine can’t keep up — offers to switch to an even lighter model. If the download fails, init still succeeds and prints how to retry. Pass --no-wait-model to return to the shell right away and let the agent finish the download in the background instead. To pin a specific model, pass --inference-bundled-model-id <id> or see Switch the bundled model.

Under-spec machines

If the machine is too small to run any usable coding model, init skips the local engine and tells you what it would need. The device still enrolls and works as part of your network — so it can use models running on your other devices, just not host one. To install the engine anyway, re-run with --inference-enabled=true.

Note

On CI or other non-interactive hosts, pass --non-interactive to accept the hardware-derived defaults without prompting. To target a non-default control plane, add --control https://your-control-plane.example.com.

Start the background service

Enrollment configures the device; the waired-agent service keeps it connected.

Linux / macOS

sudo systemctl enable --now waired-agent

(macOS uses a system launchd daemon that the installer already loaded; it starts at boot, no login required.)

Windows

Start-Service waired-agent

Next: verify it works.