Skip to main content

How It's Built

The platform is split across two distinct environments. Understanding this split is the foundation for everything else in these docs.

The Cogrion Side

Cogrion operates a shared SaaS layer that all tenants connect to:

ComponentDescription
Control PlaneThe brain. Owns the catalog of deployable applications, orchestrates what gets deployed and when, tracks the state of every resource across every tenant workspace.
CogrionThe web UI. Where tenant users interact with the platform — running SQL, building dashboards, monitoring pipelines, managing ML models.
KeycloakThe identity provider. Every login and service-to-service call goes through it. Each tenant gets their own isolated realm.
TemporalHandles long-running cloud provisioning workflows — creating VPCs, Kubernetes clusters, and IAM roles on the Cogrion side.

The Tenant Side

Everything that runs inside the tenant's own cloud account:

ComponentDescription
Cloud ProviderThe tenant's AWS, AliCloud or other cloud provider account. All infrastructure is provisioned here — the tenant owns the account, the billing, and the data.
Kubernetes ClusterA dedicated cluster provisioned by Cogrion into the tenant's cloud account. All tenant-side workloads run here.
Cluster AgentA lightweight process running inside the cluster. Continuously polls the Control Plane for work, executes it locally (Helm installs, Kubernetes manifests, Terraform runs), and reports results back. The Control Plane never opens a connection inbound — the agent always reaches out.
BFF APIRuns inside the tenant cluster. The gateway between Cogrion and the tenant's running services. Handles request routing and token exchange so Cogrion does not communicate with each service directly.
StacksThe applications deployed into the cluster — defined in the platform catalog and executed by the Cluster Agent. Examples: Trino, Airflow, Superset, Ranger, JupyterHub. See Platform Stacks for how these are authored and deployed.

How They Connect

What This Means in Practice

  • Data sovereignty — query results, pipeline outputs, and user data never leave the tenant's cloud.
  • No inbound access — Cogrion cannot directly reach into a tenant cluster. All communication is initiated by the agent.
  • Shared catalog, isolated infrastructure — all tenants share the same application catalog and control plane, but each runs on completely separate infrastructure.