Home / On-premises

On-Premises Redaction Software

Installed in your own data center, on your hardware and your network, and operated by you. The page below is written for whoever has to sign off on that, not for whoever is buying.

What Runs Where

The platform is a web application, a database, supporting services, and an AI processing tier. In an on-premises deployment all four sit inside your boundary.

The AI tier is the one that usually decides the conversation. The whole stack runs locally: language models through vLLM or Ollama, and local models for transcription, computer vision, OCR and PII detection. No content and no inference request leaves the deployment.

Infrastructure

ConcernPosition
Operating systemWindows Server; see system requirements for supported versions
DependenciesDatabase, message broker, search index and storage are required infrastructure
NetworkPorts and firewall rules are documented per service
GPUAI processing requires GPU capacity; sizing follows throughput
TLSCertificate provisioning is part of setup
PrivilegesInstallation requires administrative privileges
LicensingPer-server activation; each application instance carries its own

High Availability

The web application runs across several servers behind a load balancer, with the database on availability groups holding replicas.

The web tier is the one that scales out. Other services run as single instances per deployment, which shapes the topology you design. Plan accordingly rather than assuming every tier scales horizontally.

Upgrades and Recovery

Rolling application upgrade and zero-downtime database upgrade are supported, alongside a documented rollback and downgrade policy. Backup and restore, and disaster recovery, are part of the deployment rather than an afterthought.

Silent installer deployment is available for automated provisioning, and telemetry exports through OpenTelemetry to whatever you already run.

What On-Premises Deployment Does Not Do

  • The web application is the tier that scales out. Other services run as single instances per deployment.
  • Each application instance carries its own license activation.
  • Licensing verification requires outbound access, independently of AI processing. For a disconnected deployment it is disabled — see air-gapped.
  • Installation requires administrative privileges on the host.
  • GPU capacity is a requirement for AI processing, not an optimisation.

How On-Premises Deployment Is Evaluated

  • Installed on customer hardware in the customer network, operated by the customer.
  • The entire AI stack runs locally; no inference request leaves the deployment.
  • High availability scales the web tier behind a load balancer with database availability groups; other services are single-instance.
  • Rolling application upgrade and zero-downtime database upgrade, with a documented rollback policy.
  • Ports, firewall rules, dependencies and system requirements are documented per service.
  • Telemetry exports via OpenTelemetry.

Send Us Your Architecture Review

We will answer it against this list rather than around it.