Skip to sign-up
Why AtriumLLM

Routers see tokens.
AtriumLLM sees your business.

A basic router knows which model a request goes to and what it costs. It doesn’t know that the request carries a payroll record, a customer contract or a sales email, so it can’t treat them differently. Atrium does, and it enforces your rules before any model sees the data.

What goes wrong when the pipe is blind

Most AI traffic today moves through one undifferentiated pipe. That works for a prototype. It breaks when HR, finance and sales all start sending real business data through it.

Every payload takes the same path

An employee record and a marketing draft go to the same models, with the same logging and the same retention. Nothing in the pipe knows one is sensitive.

Policy lives in application code

Each team writes its own rules, differently, in its own app. Proving what happened means reading code, and changing a rule means redeploying.

The caller picks the model

Model choice drifts app by app. Nobody can say which model handles customer data, or swap one out without touching every caller.

Spend without context

You see cost per API key. You can’t see cost per business line, per task, or whether the spend produced anything.

One request, two outcomes

A manager asks an assistant to summarize a performance review. Here’s what happens next.

"Summarize this review for our 1:1." payload_type: hr.employee_record data_class: restricted

Through a basic router

  1. Sent to whichever model the app hard-coded, or the cheapest one available.
  2. Name, salary and rating go to the provider as written.
  3. Logged alongside everyone else’s traffic under the default retention.
  4. Billed to a shared API key.

Through Atrium

  1. Recognized as an HR record, so your HR policy applies.
  2. Routed only to an isolated endpoint. If none is configured, the request is refused, not sent somewhere else.
  3. Fields marked sensitive in your catalog are redacted before the model sees them.
  4. Traced under the retention you set, and charged to People.

Policy enforced at the door, not in the app

Every request resolves against your policy in a fixed order, from the most specific rule to the most general. Then data, latency and cost rules are checked before a model is called.

Every policy change is versioned, attributed and reversible, so you can always answer who changed what, and roll it back.

1Tenant override
2User group
3Business line and payload type
4App and task
5Task default, then global default
Then, before any model is called Data class decides isolation. Latency class picks real-time or batch endpoints. Spend caps apply per user, group and business line.

AtriumLLM fails closed

When Atrium can’t be sure a request is safe to send, it doesn’t guess. It refuses and tells the caller why.

400

No business context

Requests without their app, task and payload details are rejected. There are no anonymous calls.

400

Unknown payload type

If a payload type isn’t in your catalog, Atrium won’t infer one. Add it to the catalog, with its rules, first.

403

No safe place to send it

Restricted data with no isolated endpoint configured is refused, never rerouted to a model that isn’t allowed to see it.

What you get over a basic router

“Basic router” describes request-level routing in general. Individual products vary.
Basic router AtriumLLM
What it knowsModel, tokens, priceAlso business line, payload type, data class and who’s asking
Who picks the modelThe calling appYour registry: versioned, reversible, no app redeploys
PolicyRate limits and allowed models per keyRules by business line, data class, group and user
When unsureSends the requestRefuses and explains why
Sensitive dataPattern matching, if anyRedact or block fields by name from your payload schemas
IdentityAPI keysSSO and SCIM from your directory, with group policies
Cost reportingPer key or per modelPer business line, task, group and user, ready for chargeback
ProofRequest logsA trace ID carried into the work produced and linked to its outcome
Model typesChat LLMsLLMs plus embeddings, predictions and transcription
Model changesSwap in codePromotions gated on evaluation and human approval
Where it runsThe vendor’s shared cloudYour dedicated instance, your cloud, or fully sovereign
Commercial incentivesOften a fee on your model spendNo markup on usage, no routing bias toward any vendor

Give your AI the context your business runs on.

Atrium is in early access. Join the list and we’ll reach out as instances open up.