Cost Controls That Professionals Expect

AI usage costs behave like cloud costs: fine until the day they are not. idapt's answer is transparent, usage-based pricing with the controls in front of you, not buried in a dashboard you check after the surprise.

Before you spend: the estimate

Every message shows a cost estimate before you send it, reflecting the actual model (including what Auto routed to) and the context being sent. Long conversation? The estimate says so, which is also your nudge that compaction or a fresh chat would be cheaper.

While you spend: visible prices

Model prices are shown in the app, per million tokens for text and per unit for media (per image, per second of video, per 1k characters of speech). The models directory carries every price; comparing cost is part of comparing models, and the cheap models roundup exists because the differences are order-of-magnitude.

Bounding the spend: budgets and allowances

• Plans include a usage allowance: everyday work draws it down before anything else; the usage page shows where you stand.
• Run budgets cap what one agent run or scheduled automation may spend: the worst case is a number you chose.
• Pay-as-you-go overflow is opt-in: by default, exhausting your allowance is a wall, not a silent slide onto your Credit balance. You enable overflow explicitly, with an optional monthly cap.

After: the audit trail

The usage log records what each request cost, on which model, from which surface. When a number looks odd, you can find the exact run and its trace rather than guessing.

Good to know

• Free-tier usage has a daily allowance with no card on file: the wall is a wall, never a bill.
• Media generation shows its price in the composer before you generate; failed video generations refund automatically.
• BYOK traffic bills your provider account instead (how BYOK works).

Open your usage page after a normal week and read it like a bill: if anything there surprises you, a budget or a cheaper default model fixes it in one setting.