The waired CLI talks to the local waired-agent service. Run
waired --help, or waired <command> --help (also -h) on any command or
subcommand, for grouped, command-specific help.
| Command |
What it does |
waired init |
Enroll this device into your Waired network (Google sign-in). Run once per machine. See First run. |
waired status |
Show daemon + identity status. Add --observability for the live engine/mesh snapshot. On a service install the device state is root-owned, so run it with sudo (elevated on Windows) — without elevation it exits with a permission message instead of guessing. |
waired doctor |
Diagnose your setup with ✓/⚠/✗ checks; press f to repair anything fixable (--fix to repair non-interactively). |
waired auth status |
Show the device token state + expiry and suggest re-init if needed. Needs sudo on a service install, like waired status. |
waired logout |
Remove this device’s identity + secrets so the next init re-enrolls cleanly. |
| Command |
What it does |
waired models |
Manage local models: ls / pull / rm / refresh. ls shows the download inventory; ls --detail shows the catalog with recommended specs, hardware fit, and selection criteria (see Model catalog). |
waired runtimes |
Manage inference runtimes: ls / install / uninstall / refresh / status / benchmark. |
waired infer "<prompt>" |
Run a one-shot inference request through the Local Gateway. Add --explain for an Auto-Selector routing dry-run. |
waired inference share <on|off|status> |
Toggle whether this engine is offered to mesh peers. See below. |
waired worker get / set |
Choose where outbound inference flows: set --mode=auto|local-only|peer-preferred, or set --pin=<peer>. |
waired peers list |
List known mesh peers (DeviceID, IP, engine, GPU, model) — useful for picking a --pin target. |
| Command |
What it does |
waired link [agent] |
Set up the per-user coding-agent integration (Claude Code skills + OpenCode plugin + gateway token). See Coding agents. |
waired unlink [agent] |
Remove the integration (surgical — only undoes what link added). |
waired codeui open |
Open the bundled OpenCode coding agent in your browser, on your real project, served by your local inference. Runs as you; only you can use it (other users on the machine and the network are blocked). Useful flags: --project DIR, --no-browser (headless/SSH), --auth basic. status / url / stop manage the running instance. |
waired claude <enable|disable|status> |
Manage the Claude Code managed-settings integration (Linux/macOS/Windows): enable points Claude Code at your local inference (also done by waired init), installs the /waired-route slash command, a footer status line showing the route, and a per-turn fallback notice; disable reverts them; status shows the current state. No credential is written, so your claude.ai subscription is preserved; serving fails open to the real API when local serving is down. enable/disable need elevation (sudo, or run elevated on Windows). |
waired claude route [auto|local|anthropic] |
Show or switch where Claude Code’s requests go, live (no restart): local forces your model, anthropic uses the real API, auto is the default (local when healthy, with automatic fallback). Also available in-session as /waired-route. |
waired claude fallback [on|off] |
Show or set whether auto mode may redirect a request it can’t serve locally to the real Anthropic API. off keeps such requests strictly local (shows the error instead). |
waired claude statusline [install [--wrap]|remove] |
Manage the Claude Code footer segment that shows the current route (and, when serving locally, the model that answered the last request). enable adds it automatically (asking first if you already have a status line); install --wrap wraps an existing one; remove reverts. The bare command is what Claude Code runs each turn. |
| Command |
What it does |
waired pause |
Pause Waired routing — the local gateway stops redirecting Anthropic/OpenAI calls. Persisted across restarts. |
waired resume |
Undo pause and restore overlay routing. |
| Command |
What it does |
waired ping <peer> |
Send an overlay ping to a peer via the daemon. |
waired keygen |
Generate a WireGuard key pair (init normally does this for you). |
waired version |
Print the build version (--json for {version, buildSHA, os, arch}). |
waired update |
Check for and apply an update, staying on the host’s current channel (--check to report only, --yes to apply non-interactively, --edge / --stable to switch channel). See Install → Update. |
These apply to status / ping / models / runtimes / infer:
--mgmt <url> — Local Management API base URL (default http://127.0.0.1:9476).--gateway <url> — Local Gateway base URL for waired infer (default http://127.0.0.1:9479, the local no-token gateway; when pointing it at the token-protected http://127.0.0.1:9473, waired infer attaches the gateway token automatically if it can read it).
These two controls are independent:
waired pause / resume stops all overlay routing — both mesh routing
and the local inference gateway return 503 while paused. Use it to
temporarily take the device out of the loop.waired inference share on / off controls only whether other peers
can use your engine. With sharing off, you can still run inference yourself
(waired infer works); peers just can’t reach your model.
So on a private workstation you might keep sharing off but stay unpaused; on
a dedicated GPU box you’d turn sharing on so your other devices can use it.