Cortex works across the full stack, not just the model layer — this is the reference architecture behind every deployment.
A diagram is only useful if every layer is actually owned by someone. Here's what we design for.
Where AI meets an actual task — a request, an approval, a decision.
Coordinates multi-step tasks across tools, models, and workflows.
A controlled entry point that routes requests across model providers.
Public APIs, private deployments, or both — selected per workload.
The documents, records, and systems an AI system is allowed to see.
Who — or which agent — is allowed to do what, and under which role.
Policy, approval gates, and ongoing evaluation of output quality.
Logs, metrics, and traces across every model and agent call.
Compute, storage, and orchestration behind every layer above.
The perimeter controls that define what can reach what.