Reference Architectures

Enterprise patterns for production AI

These are opinionated, end-to-end deployment patterns built from real enterprise AI deployments. Copy the pattern that fits your stack and adapt it.

Azure OpenAI + APIM + Cognisafe

Enterprise AI gateway with OAuth2, rate limiting, and runtime security in a single request path.

AzureEnterpriseAPIM

Semantic Kernel Agent Runtime

Secure every planner step, function call, and memory retrieval in .NET and Python SK agents.

Semantic Kernel.NETAgents

MCP Tool Governance

Enforce trust, audit, and policy for every MCP server and tool invocation in your agent estate.

MCPTool SecurityGovernance

AKS Production Deployment

High-availability Cognisafe on Azure Kubernetes Service with auto-scaling, Key Vault, and full observability.

AKSKubernetesProduction

Air-gapped Enterprise Deployment

Full on-premises deployment — local models, local scoring, no internet egress.

Air-gappedOn-premisesRegulated

SIEM Integration

Feed AI threat events into Sentinel, Splunk, or Elastic as first-class security data.

SIEMSentinelSplunk

SOC Integration

Wire Cognisafe detections into your SOC workflows — PagerDuty, ServiceNow, Jira, SOAR.

SOCITSMIncident Response

Human-in-the-Loop Approvals

Require human sign-off before high-risk agent actions execute — payments, deletions, external comms.

HITLGovernanceEU AI Act

Don't see your stack?

Every reference architecture is available as a starting point. Reach out and we'll help you adapt one for your environment.