What is Roar AI
One account that gives your company AI models and somewhere to run internal apps.
Roar is two things your company gets from one account, one login and one bill:
- Every major AI model, through one key. Your team writes against a single address and can reach any model in the catalogue — without a separate account, card and contract per AI company.
- Somewhere to put the internal apps you build. Pick a repository, answer two questions, get a working web address. No servers to set up and nothing to keep running.
Most companies arrive wanting one of the two and end up using both, because the apps people build here are usually the ones that needed AI in the first place.
Which half you need
"I want to use AI in something I'm writing." Start with the Quickstart — it takes about two minutes and works with the AI libraries you already know.
"I've built something and I need it online." Start with Hosting apps. You'll do it from the console, not from code.
"I run the account and I'm deciding who gets what." Read How it fits together — it explains projects, keys and spending limits, which is the whole of that job.
What you actually get
| One catalogue | Frontier models and open-source ones side by side, each at a published price. Switching models is changing one word in your code. |
| One price per model | The rate is published before you call and holds until we change it. It does not move because of what something cost us that day. |
| One bill, in your currency | Not a dozen foreign card charges to reconcile at month end. |
| Spending you can cap | Limits per person, per project or for the whole company — as a hard stop or as a warning. |
| Apps that just run | Deploy from a repository, get a URL, add a database or your own domain when you need one. |
| Everything visible | Every request, what it cost and who made it, in one place. |
What Roar is not
Being clear about this saves a wasted evaluation:
- Not a chat product. There is no Roar chatbot for your staff to talk to. Roar is what the things your company builds run on.
- Not a model of our own. We do not train models. We serve the ones that exist, at a rate we publish.
- Not a general cloud. App hosting is deliberately narrow: internal tools and small web apps, deployed in a few clicks. If you need to design your own infrastructure, you want a different kind of provider.
- Not per-seat. You pay for what you use — tokens and app running time — not for how many people have a login.
How you'll work with it
There are two surfaces and most people only ever touch one.
The console is the website you log in to. Everything can be done here: invite people, create keys, deploy apps, set budgets, read what was spent. If you are not writing code, this is the whole product.
The API is for the code your team writes. It is compatible with the OpenAI standard, which is why existing libraries and tools work by changing a base address and a key — there is no Roar-specific library to learn.
There is also a hosted MCP server, so an AI assistant can do console work for you — create a key, check spend, deploy an app — with exactly the permissions you gave it.
Where to go next
- How it fits together — projects, keys, budgets and who can see what.
- Quickstart — your first AI request.
- Hosting apps — your first deploy.