For every request, the right choice.

AION Router puts your PGX units (NVIDIA DGX Spark) and, if you want, an external provider behind one OpenAI-compatible endpoint. For every request it decides which PGX handles it, with which model, and what may leave the company.

01

Several PGX units, one address.

Your applications call a single endpoint. The Router checks how busy each PGX is, the NVIDIA DGX Spark computer that AION On-Premise runs on, and sends the request to the least busy one. If you add another, it starts working right away, without changing your applications.

Illustrative example

02

Not every question needs the large model.

Before deciding, the Router works out how hard the request is by weighing three signals: quick rules, similarity to requests it has already seen, and a small model acting as referee. Then it picks the lightest model that can answer well.

Translate this email into English.
Fast model
Compare the contract with our draft and flag the clauses that differ.
Large model

For each team you choose what matters most: cost, quality, a balance of the two, or privacy.

03

Before it leaves, it asks you.

Inside the company the request stays as it is. If it has to go to an external provider, the Router recognises personal data such as names and emails, replaces them with placeholders and shows in the chat the text that would leave. It's up to the person who wrote it: send it redacted or keep it in-house.

Agents · chatAION Router

Compare Mario Rossi's contract with our standard draft and flag the clauses that differ.

This request can go to the external provider. It would leave like this:

Compare [PERSON_1]'s contract with our standard draft and flag the clauses that differ.

Send it redacted?

Stays in-house: PGX 1 answers with the large model.

Try choosing and see where the request goes
04

By default it stays in-house, outside only with your permission.

The external provider is an option, not a dependency. You turn it on, only for the teams and models you decide, and only towards OpenAI-compatible providers. Without this rule, every request stays on your PGX units.

This requeststays on PGX 1 · change choice

One endpoint, your rules.

The four choices follow rules you set. The AION team runs the Router console: they set up keys, rules and budgets with you and send you the reports.

Endpoint
OpenAI-compatible: the applications and chat clients you already use connect by changing only the address.
API keys
Each key sees only the models allowed for that team.
Rules
Cost, quality, balance or privacy: one rule for each team.
Budgets
A budget for each team, with costs attributed to whoever generates them.
Reports
Usage, costs and path of every request, in one view.

Want to see the Router at work?

We show you how the Router sorts your team's work across PGX units, models and, if you want, an external provider.

Office
Corso Vercelli 89, 28100 Novara

Your data is treated according to our Privacy Policy and will not be shared with third parties.