Switch the bundled model

waired init auto-selects a bundled model from your hardware profile — the largest model that fits this machine above a coding-quality floor (a capable box gets a large MoE coder; a modest laptop gets a smaller model). To override that pick — say a larger model on a 24 GB GPU, or a smaller one to save memory — set the bundled model ID explicitly. Browse the options and their requirements in the model catalog.

Tip

To pin the model at enrollment time, pass waired init --inference-bundled-model-id=<id> — that skips the hardware auto-selection. The methods below change it on an already-enrolled machine.

There are three ways to set it. When more than one is present, the higher-priority source wins:

defaults  <  agent.json  <  environment variable  <  CLI flag

1. agent.json (persistent — recommended)

Set inference.bundled_model_id in agent.json, then restart the service.

Linux

Config file: ~/.config/waired/agent.json (respects XDG_CONFIG_HOME); for the system daemon, /etc/waired/agent.json.

sudo jq '.inference.bundled_model_id = "qwen3-coder-30b-a3b-instruct"' \
  /etc/waired/agent.json | sudo tee /etc/waired/agent.json.new
sudo mv /etc/waired/agent.json.new /etc/waired/agent.json
sudo systemctl restart waired-agent

Windows

Config file: %LOCALAPPDATA%\waired\agent.json; for the system service, %PROGRAMDATA%\waired\agent.json. Edit the inference.bundled_model_id field, then:

Restart-Service waired-agent

2. Environment variable (system-wide / temporary)

Set WAIRED_INFERENCE_BUNDLED_MODEL_ID and restart the service.

Linux (systemd)

Add it to /etc/waired/agent.env:

sudo install -m 600 -D /dev/stdin /etc/waired/agent.env <<'EOF'
WAIRED_INFERENCE_BUNDLED_MODEL_ID=qwen3-coder-30b-a3b-instruct
EOF
sudo systemctl restart waired-agent

(Keep your existing WAIRED_CONTROL_URL line if the file already has one.)

Windows (service)

setx /M WAIRED_INFERENCE_BUNDLED_MODEL_ID qwen3-coder-30b-a3b-instruct
Restart-Service waired-agent

3. CLI flag (mostly for development)

Pass the flag when starting the agent directly:

waired-agent --inference-bundled-model-id=qwen3-coder-30b-a3b-instruct

Confirm the switch

After restarting, the new model is pulled on demand (this can take a few minutes):

waired models ls                # watch the download / disk usage
waired status --observability   # the Engine line should show the new model=

Tip

Switching to a model your hardware can’t fit will leave the engine not ready. Check the RAM/VRAM column for your model and runtime in the model catalog before switching.