How AI builds on Foundation

How AI consumes shared Foundation services.

Agentic Friendly

AI does not replace Foundation. It depends on Foundation to provide the secure and operational foundation on which AI application services run.

Foundation capabilities used by AI

Foundation capabilityHow AI uses it
Networking and exposurePublish the Portal, the BSQAI API, the MCP Gateway, optional sandbox app previews, and the documentation site. MLflow is internal only and is reached through the BSQAI API
Identity and accessAuthenticate users and services through Keycloak and related access controls
Secrets and certificatesProtect service credentials, client secrets, and TLS configuration
InferenceRun LLMs on KServe with platform-managed vLLM profiles, and predictive models on MLServer or NVIDIA Triton, behind the private inference gateway. KAI Scheduler places GPU workloads in each tenant's GPU queue
Tenant runtimeGive each customer organization its namespace, tenant-<slug>, with quota, LimitRange, network policies, and GPU queues, and optionally its database, object storage, private MLflow, and Grafana organization, through the BSQAI Tenant Operator
Data and storagePersist operational state in PostgreSQL, and objects in Rook-Ceph or external S3-compatible storage, and in a customer organization's own bucket where it has one
WorkflowsRun tenant lifecycle, file ingestion, reindex, Deep Research, skill publication, and managed-ingestion workflows durably on Temporal
ObservabilityCollect metrics, logs, and traces from AI services with server-authorized tenant context, and keep metrics long term in Grafana Mimir
GitOps and deliveryDeploy and update AI services through Argo CD and the platform delivery chain

Practical boundary

  • Foundation provides the shared operational platform
  • AI provides reusable AI application capabilities
  • Data and business use cases consume both

Example

When a team publishes a GenAI assistant, Foundation still handles Gateway API routes, certificates, storage, and inference operations. AI adds the portal, APIs, tenant-aware model lifecycle, vector workflows, tools, and GenAI observability that make the assistant usable.

On this page