Home / Deployment
The Same Engine, Wherever It Has to Run
Most AI redaction is a cloud service. That is a problem when the content is criminal-justice material, classified, or subject to a sovereignty requirement. Those buyers need detection that holds up and a guarantee that the content never leaves, and a hosted service can only offer the first.
Where It Can Run
Shared SaaS
VIDIZMO, multi-tenant with isolation by tenant identity
Dedicated cloud
VIDIZMO, isolated tenancy, operated by VIDIZMO
Private cloud
You, in your own cloud subscription
On-premises
You, your data centre, your hardware, operated by you
Hybrid
Cloud application reaching on-premises directories and storage
Air-gapped
You, disconnected from the public internet entirely
What Deployment Has to Answer
Does content leave?
Plenty of vendors say on-premises and mean the application runs locally while the AI still calls a hosted API. That fails the requirement it was meant to satisfy. In an air-gapped deployment the whole AI stack is local — language models through vLLM or Ollama, plus local models for transcription, computer vision, OCR and PII detection. No content and no inference request leaves.
Does it resolve in-country?
Storage, cognitive services and identity connect to sovereign regions rather than only the global one. Supported environments are Azure Global, Azure US Government, China, and German government cloud.
Can AI run without egress?
Short of full disconnection, cognitive services run in containers inside your own infrastructure, so capabilities that would otherwise call a cloud service operate without leaving the network.
Will it stay up?
A high-availability topology runs the web application across several servers behind a load balancer, with the database on availability groups holding replicas.
Can we install it our way?
Containerized deployment, bring-your-own-cloud, Azure and AWS marketplace listings, silent installer deployment, rolling application upgrade, zero-downtime database upgrade, backup and restore, disaster recovery, and telemetry export through OpenTelemetry.
What Stays the Same Everywhere
Detection classes
The full object and PII set, unchanged
Redaction styles
Blur, pixelate, solid fill, silence and beep
Exemption codes
Statutory codes and overlay text on the output
Custody trail
User, email, IP, timestamp on every action
Bulk processing
Queue-based, permissioned per format
API
The same REST endpoints and OpenAPI spec
Deploying On-Premises
Size it
GPU capacity is a requirement for AI processing, not an optimisation. Sizing follows throughput.
Provision dependencies
Database, message broker, search index and storage, with documented ports and firewall rules.
Install
Silent installer for automated provisioning, administrative privileges required on the host.
Activate
Per-server licence activation, one per application instance.
Licensing verification requires outbound access independently of AI processing, and is disabled for disconnected deployments.
FAQ
Deployment questions, answered
What deployment options are available?
Six: shared SaaS, dedicated cloud, private cloud in your own subscription, on-premises in your own data centre, hybrid where a cloud application reaches on-premises directories and storage, and fully air-gapped.
Does the AI still call a cloud service in an on-premises deployment?
Not in an air-gapped one. Plenty of vendors say on-premises and mean the application runs locally while the AI still calls a hosted API. In an air-gapped deployment the whole stack is local: language models through vLLM or Ollama, plus local models for transcription, computer vision, OCR and PII detection. No content and no inference request leaves.
Can data be kept in a particular country?
Storage, cognitive services and identity connect to sovereign regions rather than only the global one. Supported environments are Azure Global, Azure US Government, China and German government cloud.
Do capabilities differ between deployment options?
No. The full detection class set, the redaction styles, exemption codes, the custody trail, bulk processing and the REST API are the same in every posture.
What is needed to run on-premises?
GPU capacity is a requirement for AI processing rather than an optimisation, and sizing follows throughput. A database, message broker, search index and storage are provisioned with documented ports and firewall rules, installation runs through a silent installer, and licensing is activated per server.
Tell Us Your Constraint
Sovereignty, classification, or a network that does not reach the internet. The answer is usually one of the six, and we will tell you which.