Model, effort and output style

Two buttons on the left of the composer bar show which model the tab uses and how hard it thinks. They always show what the session is actually using, including changes you made with /model or another slash command.

The list of models and each model’s effort levels come from Claude Code, not from a list built into Claudette. Claude Code’s model configuration describes them.

Switch the model

Click the model, shown by its name and ▾. A menu lists the models, each with a description.
Pick a different one. A question appears, Switch model?
Choose Switch model, or Cancel.

The question warns, with the model’s name, that Switching to Haiku resets this tab’s cached context: the new model has to read the whole conversation again, so your next message uses more of your limits. If the new model doesn’t support the tab’s current effort, the question says so, and the effort goes back to that model’s default.

The conversation itself is kept. Claude Code switches the model in place, even in the middle of a turn.

Set the effort

Click the effort, such as High ▾, and pick a level. The menu offers the levels the current model supports, and Default, which leaves the choice to Claude Code. It applies from the next turn, with no question.

Ultracode

At the foot of the effort menu, Ultracode is a switch. With it on, Claude can run workflows of subagents on its own for big tasks, at the tab’s effort. It uses more of your plan, as the switch’s description says. See Claude Code’s workflows.

It applies from the next turn, is kept with the tab and is switched on again whenever the tab’s Claude Code starts. The effort button says so: High · Ultracode.

Output style

Under Ultracode, Output style, for this folder lists the styles the session offers, with the current one in bold: Default, Explanatory, Learning and any you’ve made. Pick one and it applies from Claude’s next reply. Claude Code’s output styles says what each does.

The choice is kept in the folder’s own Claude Code settings, where /output-style keeps it, so a terminal session in the same folder uses it too. A tab started without the folder’s own settings, through Start without it on the folder trust question, doesn’t offer styles, because that’s where the choice is kept (Folder trust).

A fallback model

Settings → Claude Code → Fallback model is none by default. Choose one and tabs start with it as the model to switch to when their own is overloaded, rather than waiting. It applies to tabs started after you change it.

What new tabs start with

Settings → Tabs → New tabs start with → Model and Effort set what a new tab starts with. For one tab, Tab settings… has Model and Effort (Tab settings).

Settings

Setting	Default	Where
Model	Default	Settings → Tabs → New tabs start with
Effort	Default	Settings → Tabs → New tabs start with
Fallback model	None	Settings → Claude Code

The tab’s own model and effort, and Ultracode and Output style, are on the tab, as above.