Appliance overview

The Operayde appliance is a managed environment that runs AI on your infrastructure. It is the only component in the system that ever sees a prompt or a document.

Deployment options

The appliance can be installed:

• On your premises — inside your own datacenter or server room.
• In an Operayde datacenter — hosted in EU, KSA, or UAE regions.

In both cases, prompts and documents stay within the appliance boundary and never reach the central plane.

Tiers

Starter

50–250 FTE — instruct + embedding + vision

Professional

250–500 FTE — full model zoo + custom fine-tuning

Entry

500–2,000 FTE — multi-appliance fleet + air-gapped option

Premium

2,000+ FTE — dedicated fleet with HA + multi-region

On the appliance

• A hardened Debian image with encrypted storage.
• The Operayde Workspace Runtime — the Python service that orchestrates inference, retrieval, and ingestion.
• A local vector index and document store, both encrypted at rest.
• The gateway, running as a sidecar, terminating all in/out traffic.
• A Merkle audit writer that batches events into per-day trees.
• A fleet agent that reports heartbeats, receives signed policy, and applies updates.

What the appliance does not do

• It does not connect to the internet for model inference. Ever.
• It does not execute code the user pastes into a prompt — sandboxing is strict and only approved tool-calls run, and only in ephemeral containers.
• It does not accept unsigned updates, even from the fleet.