/// node operations

wai CLI reference

The installer puts a waicommand on each node machine. Use it to check status, follow logs, move models, and manage the agent — all from the node's own terminal, no dashboard login required. Works the same on Windows, macOS, and Linux.

Looking for the API / SDK quickstart instead? See the gateway docs.

Install a node

Create a node on the Nodes page to get your wai_node_… key, then run the one-liner for your OS. No admin rights needed (Windows installs per-user).

windows · powershell
& ([scriptblock]::Create((irm https://wideareaai.com/install.ps1))) -Key wai_node_XXXX
macos / linux
curl -fsSL https://wideareaai.com/install.sh | sh -s -- --key wai_node_XXXX

Useful install flags: -ModelsDir D:\models / --models-dir /mnt/big/models to store models on another disk, and -LlamaVariant cuda|vulkan|cpu / --runtime native|container to force a build. The node auto-starts at login and keeps running in the background.

Commands

wai statusService state, health, current model, tunnel URL, media model, and a live llama-server check.
wai logsFollow the agent + llama-server logs in real time (Ctrl-C to stop).
wai startStart the node.
wai stopStop the node.
wai restartRestart the node (agent + llama-server) — e.g. after dropping a new .gguf into the models folder.
wai updateRe-install the latest agent now. Keeps your models and node key. (Also happens automatically — see Auto-update.)
wai models-dir [path]Show where models are stored, or move them to another disk. See below.
wai uninstallRemove the node from this machine entirely: service/task, agent, CLI, config (node key), logs, and downloaded models.
wai helpShow all commands. wai --help, wai -h, and wai -? work too.

Not on your PATH? Use the full path — macOS/Linux: ~/.wai/bin/wai · Windows: %LOCALAPPDATA%\wai\bin\wai.cmd

Storing models on another disk

wai models-dirwith no argument prints the current model folder. Give it a path and it moves storage there: it stops the node so files aren't locked, moves any already-downloaded models so they aren't re-fetched, repoints the config, and restarts. The folder holds all models — text plus the media subfolder for image/audio — so one command moves everything, and it persists across updates.

windows
wai models-dir              # show where models live now
wai models-dir D:\models    # move them to D:\models
macos / linux
wai models-dir                    # show current location
wai models-dir /mnt/big/wai-models  # move them there

Auto-update

Nodes keep themselves current automatically: each one checks the gateway on its heartbeat and re-runs the installer when a newer agent is available — pulling the latest agent and binaries and restarting, no wai updateneeded. Updates never interrupt a model download, and a node that can't apply an update simply stays on its current version rather than looping.

To pin a node to its installed version, set WAI_AUTO_UPDATE=0 in its config and restart. You can still update it by hand anytime with wai update.

Configuration

The installer writes a config file you can edit, then wai restart to apply: macOS/Linux ~/.wai/env, Windows %LOCALAPPDATA%\wai\env.ps1. Every value can also be passed as an environment variable.

WAI_NODE_KEYNode key (required; written by the installer).
WAI_GATEWAY_URLGateway URL (default: https://wideareaai.com).
WAI_MODELS_DIRWhere models are stored. Prefer the wai models-dir command, which also moves existing files.
WAI_MODEL_URLDefault model to download on first boot.
WAI_PORTLocal llama-server port (default: 8080).
WAI_LLAMA_ARGSExtra llama-server flags (default: -ngl 999, offload all layers to GPU).
WAI_LLAMA_CUDA_DEVICESPin llama to specific GPUs, e.g. "0,1". Multi-GPU nodes otherwise reserve the largest card for image generation.
WAI_AUTO_UPDATESet to 0 to disable automatic agent updates on this node.
WAI_MEDIA_PORTImage/audio (stable-diffusion.cpp / whisper.cpp) server port (default: 8081).
WAI_SD_ARGSExtra stable-diffusion.cpp flags for image generation.
WAI_WHISPER_ARGSExtra whisper.cpp flags for transcription.

Node states

What wai status (and the dashboard) report.

connectionhealthmeaning
ONLINEinitializingAgent just started; tunnel is up, model work hasn't begun.
ONLINEdownloading modelOne-time model download (~5GB default). Watch progress with wai logs.
ONLINEloading modelLoading weights into GPU/RAM — usually under a minute.
ONLINEreadyServing inference — the gateway routes traffic here.
ONLINEunhealthyAgent is alive but llama-server stopped responding → wai restart.
OFFLINENo heartbeat for 90s+: machine asleep, no internet, or service stopped → wai start.

Removing a node

Click [ remove ] on the dashboard, then on the machine run wai uninstall — it stops the service, removes the agent, CLI, config, logs, and models, and takes the node off your PATH. Nothing is left behind.