Concepts and architecture

Concepts and architecture#

Sovereign Data and AI Factory combines physical infrastructure, orchestration software, and lifecycle automation into a single system, preassembled and installed in your data center.

Understanding the system architecture and core roles helps clarify how Postgres workloads and AI pipelines are deployed, managed, and scaled.

As a prerequisite, see configuration details in System specifications .

Node types and roles#

The system includes several node roles, each with distinct responsibilities and configurations.

Administration node#

The administration node provides additional software packages, licensing, and support tooling for your Sovereign Data and AI Factory. You need one administration node per installation.

Hybrid Manager nodes#

Three Hybrid Manager nodes form a high-availability control plane. They manage Kubernetes scheduling, cluster lifecycle, observability, and API services.

Functions include:

  • Managing database cluster deployments

  • Monitoring 200+ metrics from Postgres, OS, and GPU workloads

  • Coordinating updates and patches

  • Serving configuration UIs and APIs

These nodes aren’t used for running data workloads.

Compute nodes#

Compute nodes are used to run Postgres databases, analytics pipelines, or vector workloads. These nodes support:

  • EDB Postgres Advanced Server or Extended Server

  • EDB Postgres Distributed (optional)

  • Lakehouse formats, such as Apache Iceberg®

  • Python-based embedding or transformation pipelines

AI nodes (optional)#

AI nodes are GPU-enabled servers that support large-scale model inference and agentic execution. The system includes two types:

  • 2-GPU nodes for mid-scale model serving (NVIDIA Ada L40S)

  • 8-GPU nodes for LLM and agent runtime (NVIDIA HGX with B200 GPUs)

These nodes integrate with KServe to provide on-premises model serving and inference routing.

Orchestration and lifecycle#

The system runs a preinstalled Kubernetes cluster managed by Hybrid Manager. You don’t directly access Kubernetes. Instead, Hybrid Manager provides a web-based interface to:

  • Deploy Postgres clusters with high availability

  • Install and monitor vector pipelines or AI workflows

  • Scale workloads to additional nodes

  • Monitor performance and resource usage

EDB provides lifecycle services for both software and hardware:

  • Major and minor version upgrades

  • Security patches

  • Hardware firmware and BIOS updates

  • Automated cluster rebalancing and restart handling

What is “sovereign”?#

Sovereign in this context means:

  • All data and models reside in your physical environment

  • No external cloud dependencies or data sharing

  • You own the hardware and control network access

  • Fully audit-ready, isolated environment for regulated industries

Sovereign Data and AI Factory is suitable for:

  • Government or national data infrastructure

  • Financial services subject to local data regulations

  • Enterprises with strict cloud restrictions or sensitive IP

Supported workload patterns#

Each configuration supports a range of use cases:

Workload | Description | Core Compute | Advanced AI ———|————-|————–|————- High availability Postgres | Run 3-node HA clusters | Check | Check Embeddings | Convert internal data into vector form | NO | Check Retrieval-augmented generation (RAG) | Combine Postgres + LLMs | NO | Check Inference | Serve models locally using GPUs | NO | Check Agentic workflows | Build AI agents that act autonomously | NO | Check

For configuration details, see System specifications .

Contents