Glossary and key terms#
These key terms and architectural components are used throughout the Sovereign AI and Data Factory documentation.
Administration node#
The central control node of the system. Hosts the Hybrid Manager interface, authentication services, and orchestration logic.
AI node#
A GPU-enabled server used for model inference, vector processing, and agentic execution. Part of the optional AI workload configuration.
Agentic workflow#
A multi-step automation pattern where AI agents perform tasks independently using internal tools, memory, and feedback loops. Often built on top of LLMs and vector data.
Control plane#
The set of three Hybrid Manager nodes responsible for cluster scheduling, monitoring, lifecycle operations, and access control. Operates independently of data workloads.
Compute node#
A general-purpose high-performance node used for running Postgres clusters, analytics, or vector pipelines. Can be scaled out as needed.
Embedding#
The process of converting structured or unstructured data into vector representations suitable for AI workflows such as retrieval and classification.
EDB Hybrid Manager#
EDB’s management layer for orchestrating database and AI workloads in Kubernetes environments. Used to deploy clusters, monitor systems, and manage lifecycle operations.
Inference#
The act of generating output from a machine-learning model. In this system, inference is performed on dedicated GPU nodes using KServe and NVIDIA NIM containers.
KServe#
An open-source model serving framework used to run inference workloads. Integrated into Hybrid Manager and GPU nodes for serving LLMs and embedding models.
Lifecycle services#
Ongoing support and management for both hardware and software, including security patches, OS updates, Postgres upgrades, and firmware management.
EDB Postgres Distributed (PGD)#
EDB’s multi-node, high-availability Postgres solution. Optional for customers who require globally distributed or synchronous multi-region clusters.
Retrieval-augmented generation (RAG)#
A workload pattern where structured data (often in Postgres) is retrieved and passed to an LLM to generate natural language responses. Requires embedding and inference capabilities.
Sovereign AI#
An AI deployment model where all data, models, and execution environments remain physically and logically within a customer-controlled space that’s fully air-gapped if needed.
Vector store#
A database or index that stores embedding vectors and supports similarity search. Postgres and pgvector are used in this system for local vector storage.