Live demo
Hybrid routing playground
Pick a request, see the gateway decide local or hosted, and read the rule that fired — the same decision that lands in the audit chain.
OpenPillar 3 · Local · Inside · Routed · Sovereign-by-default
You don't need everything on-prem — but a small set of data (M&A drafts, customer health scores, deal-flow notes, founder voice) cannot go to OpenAI, while a much larger set could happily use GPT-class models. Today you're forced to pick one model for everything and constrain based on the worst case.
Evidence
Every item below is a live demo on this site, a mechanism you can inspect in the product, or an artefact that arrives under NDA. Nothing here is a mock-up.
Live demo
Pick a request, see the gateway decide local or hosted, and read the rule that fired — the same decision that lands in the audit chain.
OpenInteractive
From Starter to Premium. Concurrency, throughput, hardware BOM range — all in one table, one click away.
OpenPolicy
Tag your data once. Routing rules respect the tag forever. Confidential content stays local while general queries use frontier APIs.
OpenDocs
Llama 3.3 70B, Qwen 2.5/3 72B, Mistral Large, DeepSeek V3 — all hosted on Operayde appliances out of the box. No vendor lock-in at the model layer.
OpenDiagram
EU-West, EU-South, EU-Central, GCC sovereign — pick where each tenant's compute lives. Routing rules respect the residency boundary automatically.
OpenHow it works
Operayde's appliance hosts open-weight models on your hardware. The gateway routes prompts based on policy: sensitive data + sensitive operation = local; everything else can use whichever frontier provider you allow. Both branches emit into the same audit chain so you can prove provenance for any output. Routing rules are yours — you read them, you edit them, you version them. We don't hide them in a 'smart' black box.
Read the full system viewWhat it costs
Local hosting starts at the Starter tier. On-prem appliances scale from a deskside unit to a Premium cluster. The discovery agent picks the right tier based on your concurrency, model size, and residency posture.