Deployment Models

Three deployment options: SaaS (Free and Sovereign), Sovereign dedicated infrastructure, and Enterprise SDK with air-gapped Virtual PLC Docker.

Deployment Overview

Model	Plans	Compute	Data Location
SaaS (Shared)	Free	Shared capacity	RadMah managed cloud — your residency region
SaaS (Priority)	Sovereign	Priority capacity	RadMah managed cloud — your residency region
Enterprise SDK	Enterprise	Customer infrastructure	Customer-controlled

SaaS Deployment (Free and Sovereign)

The standard deployment for Free and Sovereign plans. RadMah AI manages all infrastructure in the residency region selected for the Account. No customer infrastructure required. The European Union plane is the region currently serving production; the United States plane is defined and restorable but is not deployed at present, so a US residency cannot be placed today.

Infrastructure Stack

API: FastAPI on managed containers
Workers: Celery on managed containers, across queue tiers
Database: managed PostgreSQL with point-in-time recovery. Zone-redundant high availability is a capability of the deployment, not currently enabled on the serving plane
Cache/Broker: managed cache and queue, provisioned separately
Storage: managed object storage for artifacts and evidence bundles
GPU Jobs: GPU worker tiers are defined in infrastructure; the production fleet runs on CPU today
Secrets: the platform key vault, resolved at runtime
Identity: managed workload identities, no static credentials
Frontend: managed container app

ℹSovereign Queue Priority

Sovereign plan jobs run with priority over Free-tier jobs, which run on spare capacity. Sovereign concurrent job limit: 10 (vs. 1 for Free).

Enterprise SDK Deployment

Enterprise customers deploy the RadMah AI SDK and Virtual PLC Docker containers on their own infrastructure, so generation and storage stay in the customer environment.

✦Complete Data Sovereignty

All generation, training, evidence production, and verification happen entirely within the customer environment. No telemetry or data is sent to RadMah AI.

Components Deployed On-Premise

RadMah AI SDK: Python SDK package (same as SaaS)
Virtual PLC Docker image: Container-isolated Virtual PLC simulation runtime
Cryptographic core module: Signed native module that performs evidence sealing and verification locally
Engine core: Full generation pipeline runtime for all supported engines

Enterprise Virtual PLC Runtime Requirements

Real-time-capable Linux container with Docker installed
Minimum 4 CPU cores, 8 GB RAM per PLC instance
Max 50 concurrent PLC instances per deployment

Air-Gapped Deployments

ℹFully Offline Operation

Enterprise SDK supports fully air-gapped operation. Evidence bundle verification is self-contained — all hashing and parsing runs locally inside the cryptographic core module, with no network calls required. See Hybrid SDK for air-gapped configuration.

Infrastructure as Code

SaaS infrastructure is fully managed via Terraform. The IaC covers:

Virtual network, subnets and private endpoints
Container platform, app definitions and services
Managed PostgreSQL with automated backups and point-in-time recovery
Managed cache and queue, provisioned separately
Managed object storage for artifacts and state
Compute job pools per tier
Least-privilege roles bound to managed workload identities
Per-region log workspace, metrics and alert rules
DNS zone records
Managed container app deployments

⚠Region Policy

RadMah AI is a dual-residency product: the European Union and United States planes are each defined with their own network, database, storage, cache, key vault and logging, under consistent naming, tagging and description policies. A request that would be served from outside the promised jurisdiction is refused rather than redirected. One plane — the European Union — is deployed today. The jurisdiction controls, the residency selection and the refusal behaviour are the same in either configuration; what changes is how many planes are switched on.

SLA

Plan	SLA
Free	None
Sovereign	99.9% uptime
Enterprise	Custom (negotiated)