Two controls shape the assistant’s answers: the model you pick and its thinking level. Behind both is a monthly AI budget that every workspace draws on. This page covers the controls and the budget; for what the assistant does with a draft, see The AI assistant.
Choosing a model
The assistant has a model picker. Models come in two groups: a standard set on every plan, and premium models on paid plans. The default is a fast model available everywhere, and you can switch at any time.
Model access follows the workspace, not the person. In a paid workspace (Craft or Craft Team), the picker shows every model. In a free workspace (Open), premium models are locked and marked with an upgrade prompt, and you choose from the standard set. The same account sees every model in one workspace and only the standard set in another, depending on each workspace’s plan.
Thinking
Thinking controls how much the model works through a problem before it answers, with four levels: Off, Low, Medium, and High. Higher levels use more of the budget, and can produce better answers on harder problems.
Some models always reason at a fixed level. For those, the picker shows that default in place of the control, and you cannot change it.
Thinking is set per model: switching models shows that model’s level, and a new conversation starts from your default.
Setting your defaults
Your default model and thinking level live in your AI preferences, alongside the choice of which models appear in your picker. Those defaults apply to new conversations.
You can also change the model or thinking level inside a conversation. That choice sticks to that conversation, without changing your defaults. Your AI preferences are personal, and you reach them from Settings.
The monthly AI budget
Every workspace has a monthly AI budget that the assistant draws on. It is shared across the whole workspace (everyone’s assistant use comes from the same budget), and it resets each month.
Usage is measured by the cost of each request, which depends on the model and the thinking level. A premium model at the High thinking level uses more of the budget than the default model with thinking Off. Paid plans include a larger monthly budget than the free plan.
The workspace owner sees usage in workspace settings under AI & Usage:
| What you see | Detail |
|---|---|
| Budget used | How much of the monthly budget is spent so far |
| Reset date | When the budget refills for the next month |
| Members who used the assistant | Which members have run the assistant |
| Per-member spend | How much of the budget each member has used |
Members who are not the owner see a note that the owner manages usage, in place of the figures.
AI credits
On paid plans, the owner can add AI credits for assistant use beyond the monthly budget. Credits cover the gap once the monthly budget is spent: they are used only after the budget runs out, never before.
- Credits do not expire while the workspace stays on a paid plan.
- Unused credits can be refunded within a limited window after purchase.
- Credits do not unlock premium models. Model access is set by the plan, not by credits.
AI credits are not available on the free plan. For what each plan includes, see Plans & limits.
When the budget runs out
When the monthly budget is spent and no credits remain, the assistant pauses for the whole workspace until the budget resets or more capacity is added. A banner states the pause and offers actions for your plan: on a free workspace, change the plan; on a paid workspace, add credits or change the plan.
The pause stops new requests. It does not touch what is already there: conversations stay readable, and any proposals already made can still be accepted or rejected.
Related
- The AI assistant: edit proposals, attachments, and web research
- Plans & limits: what each plan includes, and where to manage billing
- Settings: your personal model and thinking defaults