Set centre budgets and stop unexpected spending

For: Student, Teacher, Centre Manager, Super Admin

Look for: AI & spending

1. For a quick change, use Budgets & top-ups at the top. Add an expiring allowance for today, or open Edit all student baselines for a table of recurring defaults and individual overrides. The advanced AI & spending screen still controls models and other AI settings.
2. Manager/admin: open AI & spending in the correct centre. Amount fields use US cents: 100 means $1, 25 means $0.25 and 2,000 means $20.
3. Set the shared centre AI + voice + image budget and the shared child budget. Then set text, voice and image budgets for the centre and child. Every applicable limit must allow a request.
4. Set operation counts too: overall requests, text replies, voice operations, recorded-input seconds, new-image counts, live minutes/day and minutes/call. Raising money alone does not raise these caps.
5. Set active learning minutes and the learning-day time zone separately. Decide whether teachers may extend time, grant live voice or raise child image counts; teachers request top-ups or apply them within a Super Admin-approved allowance. They cannot change permanent budgets or the Super Admin ceiling.
6. Budget content/audio preparation and rechecks separately. Select tutor, checker and live models, reasoning and output limits. Keep the pricing review reference and expiry current.
7. Use AI cost insights to plan the number of children, minutes and speaking pattern. Allow room for simultaneous pending replies, then monitor settled estimates and held amounts during a small trial.
8. Provider account credits are separate from app allowances. If credits/key errors paused the provider, fix the cause and use Resume after fixing credits or key. Saving a larger budget does not buy provider credit.
9. Press Save limits and read any validation message. Spending and operation counts reset at midnight UTC (5:30 am in India); active learning time follows the selected learning-day time zone instead.

Reasoning settings are separate for the text tutor, correctness checker and live voice. Select a model-compatible effort; GPT-4.1 uses none. Text output limits include reasoning, so allow at least 2,000 output tokens when reasoning is enabled. More reasoning can cost more and take longer; it does not guarantee a correct answer. Check the learner's maths and language experience before changing the centre default. Image detail controls rendering, not the checker's reasoning.

This guide explains the feature; reading it does not change any setting or grant access. Back to all help