Ask a finance team what their company spends on AI and you will usually get a shrug. One department pays for an assistant. Another signed up for something else. A third put a subscription on a personal card and expensed it. Nobody owns the total, nobody can say which tools saw customer data, and the person who could answer left in March.
We built Roar AI Cloud around the opposite arrangement: one key for every model, one bill, and one place to see what happened.
Every model, one key
Your team gets an endpoint that reaches the frontier models and the open-source ones — chat, embeddings, images, speech and transcription. Everyone uses the same key. When something better ships, it is already there.
One thing we do differently, and it is the part we care most about. Most gateways pick who serves each request on price, per call — which is why the same model can accept a setting one minute and reject it the next. We choose the route once and hold it. What you tested is what ships.
The price holds too. A published rate is stored against the model, so a provider raising theirs costs us margin rather than reaching your invoice mid-month. You can read every rate before you spend anything.
Somewhere to put what you build
The interesting AI work in a company rarely starts in the engineering team. It starts with the person who knows the problem — the recruiter who wants résumés summarised, the ops lead who needs a small internal dashboard.
Those people can now build the thing and put it somewhere. Connect a repository, and you get a live URL with a certificate, a Postgres database of its own, scheduled jobs, and outbound email working on the first deploy. No servers, no ticket, no waiting for a platform team that has other priorities.
The database sits in the same region as the app, which is less of a detail than it sounds: we measured the round trip at 0.31ms in-region against about 35ms to the next region along. A page render is five to twenty queries, so that difference is the difference between an app that feels instant and one that does not.
Three regions, and one of them is here
Colombo, Singapore and Bangalore are serving today. Pick the one your users are in, or the one your data has to stay in.
Colombo is the reason the company exists. Sri Lankan teams have spent years sending their traffic — and their customers' data — to a datacentre on the other side of an ocean, paying for the latency and the exchange rate both. That is a solvable problem and we are solving it.
Controls, because this only works if IT says yes
- Budgets on a key, a project or a team, counted from real usage. The tightest cap wins and a request over the line is refused.
- Guardrails that scan for personal data and secrets before a request reaches an external model — redact it, block it, or send that call to a private one instead.
- Scoped keys, so the one in somebody's prototype cannot reach production.
- Your call history, encrypted at rest, and switchable off entirely.
- Billing in rupees, on one invoice.
Where we are
Early access, deliberately small. We would rather bring on a handful of teams and fix what they run into than open the doors and apologise later.
If you have something you want to run, request access and tell us what it is.