Using the workspace

The model picker

Choose which model answers in Duscode, from the top bar. Auto picks the cheapest model that can do the task, or you pin a model, with tags, a cost hint and its effort. Open it with Alt+M.

The model picker is in the top bar, on the right. It reads Model · Auto, or the name of the model you pinned. It sits apart from the message box's mode control, which has an Auto of its own: that one is about how much Dus may do without asking, this one about which model answers. Click it, or press Alt+M, to open it.

Auto

Auto is the default and the first item: Cheapest model that can do it. Each task starts on a fast model, and Duscode moves it up a tier when the checks fail after a fix, when Dus gets stuck, or when the model refuses. Once there's enough history in a project, Auto starts each kind of work (bug fixes, features, docs…) on the model that has finished it most cheaply there. A provider that is out of credit is skipped until it has credit again. See Routing and costs.

While Auto is on, the button shows the model answering now, for example Auto · GPT-6.1 Sol. The chat's route line says why that model was chosen. Dus knows it too: ask which model are you using?, even while it works, and it tells you the model, its tier and why.

When routing has saved money in the last seven days, Auto says how much, for example saved $4.20 this week. That's measured against running the same work on the strongest model.

Pinning a model

Below Auto are the models you have a key for, strongest tier first and cheaper first within a tier. Each shows:

• its name, such as Sonnet 5.5
• muted tags: the context window (1M, 272K), its reasoning effort (Low, Medium, High) and Vision if it reads images
• a cost hint: $, $$ or $$$, which shows where its price sits among your models
• ✓ on the one in use

Click a model to pin it for this session. A pinned model answers every task in the session. It's never moved up a tier or swapped for a fallback. Choose Auto again to go back to routing.

If you change the model while a task runs, the change applies from the next task.

Effort

Models with a reasoning effort setting have a submenu. Hover the model or press → to open it, then choose Low, Medium, High or Max effort. The model's own default is marked default. Choosing an effort pins the model with that effort. The effort then shows on the picker's button and is saved with the session.

Higher effort lets the model think longer on hard problems, and it costs more per task.

Add models

Add models is at the bottom. It opens Settings → Models & keys and says how many more models a key would add, for example 6 more with a key. Add a key there and its models appear at once. See Add a provider key. See Providers and models.

Keyboard

In the terminal app, use /model <id> and /model auto.