Why we ship a box

The AI industry spent 2023–2025 convincing itself that every company should pipe prompts to someone else’s GPU cluster. Most companies quietly disagreed. They wanted the capability, but not the egress.

We built Operayde for those customers.

The pitch in three sentences

1. A hardened appliance runs on your network and does the inference.
2. A gateway on the appliance authenticates every request and emits a Merkle-signed audit event before the model sees a token.
3. A central plane runs the fleet without ever seeing your data.

Why a box and not a container

Because the boundary customers care about is a physical one. A container can be exfiltrated, migrated, copied. A box that sits in your rack, signed by us, attested by its TPM, and unable to negotiate anything other than the outbound endpoints we’ve baked in — that’s a boundary your security team recognises.

Why a fleet and not standalone

Because security patches, model updates, and policy rotations need to happen continuously. Asking customers to manage their own AI stack is how shadow IT wins. We manage the stack; you keep the data.

What we won’t do

We won’t build a public cloud inference gateway. We won’t make a chatbot that magically “uses your documents” via an opaque cloud RAG. We won’t offer a “private mode” on a shared backend and pretend the word means what it used to.

The right answer for companies with privacy obligations is simple, boring, and unexciting at conference demos: ship a box, run it, and don’t be in the data’s critical path.