Documentation
FAQ
The questions customers ask most often. If yours is not here, the page it belongs to probably answers it — and if not, write to us.
Can I sign up myself?
No — Kakehashi is invite-only. An operator invites your email address, or an admin at a company already using Kakehashi invites you into their organization. Write to the address at the bottom of this page to start that conversation.
I lost my API key. Can you send it again?
No, and that is deliberate: a key is shown once, at creation, and never stored anywhere it could be read back — not by you, not by us. Revoke the lost key from the portal and issue a new one.
How many keys can I have?
As many as you find useful. A key per application or per environment is a good habit: it makes spending legible and lets you revoke one without disturbing the rest.
Can I restrict a key to one model?
Not today. Every key reaches every model in the catalogue. A budget cap on the key is the control available, and it limits how much the key can spend rather than what it can call.
Which model should I use?
Start with a mid-sized one, measure, then move. Larger models cost more per token and are not better at everything, so the cheapest model that does the job is usually the right default. Ids and context windows are on Models.
Will my existing OpenAI code work?
If it calls an OpenAI-compatible endpoint, yes — change the base URL and the key, and pick a model id from our catalogue. See API reference.
Does my credit expire?
No. The balance does not reset on a monthly cycle and is not consumed by time — only by requests.
Can it top up automatically when the balance runs low?
Not yet. Top-ups are made by hand from the portal. Threshold alerts exist so you are warned in time to make one — see Notifications.
Can I change my billing country?
Not from the portal: the currency and the tax treatment both follow from it, so it is fixed when the account is created. Contact support if it is wrong.
Does each colleague need their own credit?
No. An organization has one shared balance that every member's keys draw on, and group limits are how you stop one team spending all of it. See Organizations and groups.
Can I be charged more than I topped up?
No. This is a prepaid service: once the balance is used up, requests are refused rather than run on account.
A model returned an empty answer. Is something broken?
Check finish_reason. If it is "length", a reasoning model spent the whole max_tokens allowance thinking before writing anything. Raise the limit and ask again.
Something is failing and I do not know why.
Start with Errors and limits — the status code usually names the cause. If it does not, send us the details listed there and we will trace it.
Still stuck? Write to kakehashi@classmethod.my.