One balance for the org, one invoice for finance
A household of four, a two-person studio and a ten-person agency run the same mechanic at different seat counts: shared prepaid credits instead of a subscription per person, roles so the person who pays is not forced to read the work, and a quota per person so no one can spend the month.
- One shared balance
- Five roles
- Quotas per person
- Runs in europe-west1
What finance and whoever runs IT actually get
Six things that exist today, none of which needs a procurement cycle to switch on.
One balance for everyone
Every member draws from the same prepaid credits. No seat licences, and nothing to reconcile across a subscription per person per tool.
Five roles
Owner, admin, developer, billing and viewer. A billing-only member sees the invoice and never the work — which is usually the whole reason someone hesitates to share an account.
A quota per person, not just per key
Named quota presets attach to a member as well as to an API key: which models they may reach, requests per minute, and a budget per period. One person cannot spend the month.
Private per person, shared where it helps
Everyone gets their own login and their own chat history. Sharing is a choice per chat — keep it private, share with the workspace, or invite named people to read or write.
Every modality included
Chat, image, video, voice, music and 3D on the same balance, so nobody has a reason to put a personal side subscription on the company card.
Custom allowance and terms
A credit allowance sized to your real usage, invoicing in your currency, purchase-order billing and payment terms. Contract rather than engineering, so it is negotiable.
The limit sits where the spending happens
Governing an AI budget by policy document does not work; governing it at the credential does. A quota preset on a member or on a key carries its own allowed models, rate limit and budget period, so production and experimentation cannot reach into each other's month.
- Quota presets attach to a member and to an API key, and are named so they are reusable
- Allowed models per preset — a member reaches what you decided, not the whole catalogue
- A credit threshold per period, notifying when the workspace crosses it
- A fallback chain, set once per workspace and model, already covers every credential pulling from it — no need to repeat it when a provider has an outage
Frequently asked questions
How many people can share one workspace?
As many as you invite. Everyone gets their own login and their own chat history; the credits and the invoice stay shared.
Can we cap what one person spends?
Yes, and per person rather than only per key: a named quota preset sets which models they may reach, requests per minute and a budget per period. You can also set a credit threshold for the whole workspace, which notifies when it is crossed — it notifies rather than pauses, because being cut off mid-run is worse than knowing.
Can we pay by invoice against a PO?
Yes. Credit allowance, invoicing in your currency, purchase-order billing and payment terms are all negotiable. Unused credits stay usable to the end of the period, and there is no cancellation fee.
Do you have SOC 2 or ISO 27001?
No. Neither is in place, and we will not claim one before it is — an attestation you can verify is the wrong thing to be optimistic about. What you can have today is the published subprocessor list, where our infrastructure runs, and how data is handled.
Do you support SAML SSO?
Not yet — no SAML, and no SCIM provisioning either. Sign-in is email, or OAuth with Google, GitHub and Microsoft. If either is a hard requirement, say so early and we will tell you straight away whether we can commit to building it.
Where does our data go?
Our own infrastructure — database, cache and object storage — runs in Google Cloud europe-west1. Inference is served by the provider whose model you choose, so those requests leave for that provider; the subprocessor page names each one and where it sits. We train nothing ourselves — there is no training pipeline in the product — and what an upstream may do with a request is governed by that provider’s own terms, which is why we name them rather than paraphrase them.
Is it suitable for a family, including kids?
The same mechanic works: separate logins, private histories, and a quota preset per person that limits which models they can reach. What it is not is a supervised or filtered product — you are choosing which models a member may use, not delegating judgement to us.
Can we cancel?
In one click from settings, any time. Remaining credits stay usable to the end of the period, and everything you generated stays in the workspace library to download or delete.
Tell us what your review needs
If there is a requirement we cannot meet, you will hear it on the first call rather than in week four. Free trial credits in the meantime, no card.