Choose which AI model runs

Waired picks a model for you when you sign in — the best one this computer can actually run. You only need this page if you want to change it.

Should you change it?

If you want…	Do this
Better answers, and you have memory to spare	Switch up a model
Faster answers, or your computer feels slow	Switch down a model
To know what you’re running right now	See Which model am I running? below
Nothing in particular	Leave it — the automatic pick is a good default

Bigger is not always better

A model that does not fit in memory does not run slower — it fails to load, and your AI stops answering. Check the memory column for your model in the model catalog before switching up.

Which model am I running?

Click the Waired icon — in the menu bar on macOS, next to the clock on Windows and Linux → Inference. The active model is listed there.

From a terminal

waired models ls              # what is downloaded, and what is active
waired models ls --detail     # every model in the catalog, with what fits this computer

Switch models

From the Waired icon

1. Click the Waired icon (menu bar on macOS, next to the clock on Windows and Linux).
2. Open Models.
3. Pick a model from the list.

The change applies live — the AI reloads in a few seconds and keeps serving. If the model has not been downloaded yet, the download starts and the icon shows progress; the old model keeps answering until the new one is ready.

After a speed test

waired runtimes benchmark

This measures how fast this computer actually is. If a different model would suit it better, Waired offers the swap and names both models with their quality tier, so you can weigh speed against quality. Accept and the download runs in the foreground with a progress bar — press Enter at any time to let it finish in the background instead.

On a server with no desktop

A headless machine has no Waired icon, so set the model in the config file and restart the background service.

Linux

Edit /etc/waired/agent.json and set inference.bundled_model_id to the model ID you want, then:

sudo systemctl restart waired-agent

Windows

Edit %PROGRAMDATA%\waired\agent.json the same way, then:

Restart-Service waired-agent

Model IDs are the values in the ID column of the model catalog, e.g. qwen3-coder-30b-a3b-instruct.

Other ways to set the same thing

The environment variable WAIRED_INFERENCE_BUNDLED_MODEL_ID and the --inference-bundled-model-id flag also work, and override the config file in that order. See Advanced install options.

Confirm the switch

waired status --observability

The Engine line names the model that is loaded now:

Engine:   ready (model=qwen3-coder-30b-a3b-instruct, 0/4 slots used, inflight=0)

If it says not ready, the model is probably still downloading — waired models ls shows how far along it is. If it stays not ready after the download finishes, the model does not fit this computer: switch back down.

Waired will let you pick a model that's too big

It warns and asks you to confirm, showing the shortfall (e.g. needs 32 GB RAM (have 31 GB)) — it does not block you. Marginal over-spec usually works and is just slower.

Free up disk space

Models are several gigabytes each. To delete one you are not using:

waired models rm <model-id>

You will be asked to confirm first.

Model catalogEvery bundled model, with the memory it needs and how good it is.

How Waired picks for youThe rule behind the automatic choice.