ATLAS · Sovereign AI gateway
One gateway between your organisation and every AI provider. It ends shadow AI, decides which providers see your prompts, and checks, routes and audits every request.
One control plane for
Cloud, EU-hosted or on-premises. Add a provider in minutes and every team can use it, under the same rules.
The problem
Two promises
Shadow AI
Staff use whatever AI they can find, with company data and no oversight. ATLAS gives them a sanctioned route that is easier than the shadow one: every approved model, one key, their own sign-in. Shadow use moves onto the record, attributed to a person, team or project.
Sovereign AI
Not the vendor, not the developer. ATLAS puts every provider behind your rules, so sensitive prompts only reach the models and regions you approve.
Want to know where your AI traffic goes today? Book a meeting
How it works
Applications point their existing OpenAI SDK at ATLAS. Nothing else changes, and nothing gets past the gateway.
01
A personal or service API key, tied to a user who signs in through your own identity provider.
02
Is this model allowed for this organisation, space and project?
03
Spending limits are checked before any tokens are used.
04
The request goes to the right provider, streamed back in the API format your app already speaks.
05
Who, which model, tokens, cost and outcome are logged for every request.
Governance built in
Model your organisation as divisions, units and teams, and run projects alongside them. Every level sets its own budget and model allow-list. A team can never spend more than its parent allows.
Projects run alongside, with their own budget
Team · Claims is close to its limit. Requests are refused once the budget is spent.
Users sign in with their own identity provider, such as Microsoft Entra ID. No separate passwords.
Personal and service keys with expiry and suffix tracking. Revoked keys stop working on the next request.
Every request and conversation, with user, model, tokens, cost and outcome.
Cost, tokens and savings from caching, ranked by space, project, user and model.
Allow EU-region or self-hosted models for sensitive work. Prompts only reach providers you approve.
Why it matters
Every team, project and model shows exactly what it costs, in euros, before the invoice arrives. Finance gets facts, not estimates.
of AI requests attributed to a person, team or project
industry-standard APIs spoken: OpenAI and Anthropic
checks on every request before a single token is used
settings to change in existing apps: base URL and key
Spend per month against your budget, with the money saved by caching shown alongside.
Saved by caching
78%
Cost per space, project, user and model, converted to euros at ECB rates. Finance gets a recharge model, not a surprise.
Provider credentials are stored encrypted inside ATLAS. Staff never see them, so they cannot paste them anywhere.
When a provider rejects or renames a parameter, ATLAS learns the fix and applies it to future requests. Fewer failed calls without anyone stepping in.
Apps speak the OpenAI or Anthropic API. ATLAS routes them to any provider, so switching vendor never means rewriting code.
See ATLAS with your own providers and teams. Book a meeting
Two lines of code
Point any OpenAI- or Anthropic-compatible client at ATLAS. Your teams keep their tools. You get control, visibility and one bill you can explain.
client = OpenAI(
base_url="https://atlas.your-company.eu/v1",
api_key=ATLAS_KEY,
)Enterprise AI gateway · by ilionx