Your AI spend — per-model breakdown and month-end projection

Every AI call Wally makes on your behalf is priced and recorded the moment it happens — metered at the rail — so your AI cost is a running line item, never a surprise at the end of the month. Settings → Usage → "Model mix & projection" shows where that spend is going.

What the card shows:
• Model mix — how your spend splits across models this month. Turtini runs two Claude models: a high-performance one (Sonnet) for quality-sensitive work like Design Studio layouts, and a low-cost one (Haiku) for high-volume work like chat and email drafting. The bar shows the share of spend each took.
• Per-model detail — for each model, the cost, the number of turns, and the input/output tokens it used.
• Month to date — your total Wally spend so far this month.
• Projected month-end — a simple straight-line forecast of where you'll land if your current daily pace holds. It appears from the second day of the month onward.

Why two models?
Matching the model to the task is how you keep AI cheap without giving up quality: heavy design and vision work goes to the strong model; everyday chat and drafting go to the low-cost one. You get the quality where it matters and the savings everywhere else.

Reading the numbers:
The per-model split always adds up to your total Wally charge for the month — it's the same money, just broken out by model. Cost figures already include the standard +15% platform chargeback, so what you see is what you're billed.

Older months:
Months before per-model tracking was turned on show the combined Wally total only, without the split. New months fill in the breakdown automatically as Wally runs.

Want to cap it?
See "Setting an AI spend guardrail (monthly budget)" to put a monthly ceiling on AI spend that automatically protects your budget without stopping your work.