Our models. Our gateway. Your AI. — the two layers under everything we build

Most AI products are a thin wrapper over someone else’s model API. Arpanet is the opposite. We build the layer underneath — the models and the gateway — and then build products on top of it. That is what “Our models. Our gateway. Your AI.” means.

The bottom layer: five in-house model families

Underneath everything sit five model families we train and serve ourselves: Verinova for language, Blab for text-to-speech, Solab for speech-to-text, Spook for vision and OCR, and VeriRAG for embeddings and reranking. Language, speech, voice, vision and retrieval — the full set a real product needs, tuned in-house for production rather than rented per call.

The gateway: one OpenAI-compatible API

Qevron is the control plane. Point your code at a single OpenAI-compatible endpoint and reach our first-party models plus 43+ external providers, with routing, caching, monitoring and cost analytics in one place. You are never locked to one vendor, and you can see every token.

The top layer: seven products

On top of that platform sit seven products — Qevron (gateway), Calleague (voice agents), Blab (voice studio), Talky (communications), SetGet (work OS), Sosyabot (social studio) and Pickerman (media intelligence). Each draws on the same models and the same gateway, so nothing is a black box bolted on from outside.

Our models. Our gateway. Your AI.

Why the integration matters

Owning the layers underneath changes what we can promise on top. You can run the whole stack on-prem, fully isolated, or in the cloud — you decide where your workloads and data live. Sovereignty is a setting, not an upsell. And because the models are ours, there is no rented intelligence and no vendor lock-in baked into the foundation.

That is the thread through everything we publish here: the platform underneath, the products on top, and the choices that ownership gives you.