Skip to main content
Operayde
Talk to us
Product

Four components, one coherent system.

The Operayde platform is made of four surfaces — an appliance, a gateway, an operator portal, and a central plane — that snap together into a single managed service. Each surface has a single job.

Component
Appliance
Your inference engine, on your site.

A hardened Debian server with open-weight models, retrieval pipelines, and the evaluation harness. It runs every workload locally and writes every event to a Merkle-signed audit chain. The hardware is leased — ownership transfers after 36 months.

Component
Gateway
The only door to the outside world.

An OpenAI-compatible API layer that handles virtual-key auth, OPA policy enforcement, model routing (local or cloud), request classification, and per-key usage metering. Every call passes through the gateway — nothing reaches the model without a policy decision.

Component
Operator Portal
Your ops console.

A single-page app for your IT and compliance teams. Manage keys, review audit trails, set budgets, view model usage, configure retrieval, and push policy updates. The portal reads from the central plane — it never touches customer data directly.

Component
Central Plane
The fleet brain.

A cloud-hosted control surface that manages fleet identity, OTA updates, billing aggregation, config distribution, and certificate rotation. It sees metadata only: hardware health, policy versions, usage counters. Prompts and documents never reach it.

Pick the appliance. We handle the rest.

Hardened hardware, signed software, fleet management, and the audit pipeline — bundled into one monthly price. No GPU procurement. No MLOps hires. No cloud bill surprises.

Recommended
T1

Starter

20–250 users

Technical spec
InferencevLLM (Paged Attention)
AcceleratorNVIDIA L40S / Pro
  • Fits under a desk or in a 1U rack slot
  • 7B–34B open-weight models
  • Private document Q&A from day one
  • Fleet managed remotely — no IT overhead
Talk to us
T2

Professional

150–500 users

Technical spec
InferencevLLM (Paged Attention)
AcceleratorNVIDIA L40S / H100
  • Dual-GPU for 70B-class reasoning models
  • Fine-tuning pipeline on your data
  • Dedicated deployment engineer
  • Custom eval benchmarks included
Talk to us
T3

Entry

500–2,000 users

Technical spec
InferencevLLM (Clustered)
AcceleratorNVIDIA H100 / A100
  • Multi-appliance fleet, single control plane
  • Air-gapped deployment option
  • Compliance evidence pack for auditors
  • 24/7 incident response included
Talk to us
T4

Premium

2,000+ users

Technical spec
InferencevLLM (Clustered HA)
AcceleratorNVIDIA H100 / H200
  • Dedicated fleet with high availability
  • Multi-region active-active topology
  • Named account engineer on speed-dial
  • 4-hour on-site parts replacement
Talk to us
Architecture

Three surfaces, one clean contract.

The appliance lives in your building. The gateway is its only door to the outside world. The control plane never sees your data — only the metadata it needs to run the fleet, bill usage, and verify audit trails.

Operayde Reference Architecture v0.2Detailed view of the Operayde managed AI platform, showing the multi-tenant central plane, the secure mTLS channel, and the internal components of the customer-side appliance.Central PlaneFleet ManagementKnowledge DistributionSigned Update ServiceAI Models GatewayProxied & Metered LLMIdentity & Auth BrokermTLSTunnelCustomer Appliance (On-Premise)Policy Engine (Ingress / Egress Guardrails)Client-facing APILLM Router (LiteLLM)A. WorkingB. EpisodicC. SemanticD. DocumentalE. ExternalMemory Orchestration Subsystem (Hierarchical & Permit-Aware)LLM Supply (Inference Layer)Local Inference (On-Box)Ollama (SME) or vLLM (Ent)Hosted via GatewayAzure · Anthropic · OpenAIGovernanceObservability (Langfuse)Merkle-Signed AuditPolicy Store (OPA)Secrets & KMSFleet MonitoringUser Browsers / CLICompany Data SourcesSlack / Teams
Security Boundaries
Memory Hierarchy
Managed Gateway
Local Inference

Ready to put AI behind your own firewall?

Spend 20 minutes with one of our deployment engineers. We'll walk through your workload, pick the right tier, and ship an appliance to your office within two weeks.

Product · Operayde