Sovereign AI Explained ====================== **Sovereign AI** is a core principle of **EDB Postgres® AI** (EDB PG AI): Giving you full control over your **data**, **models**, and **AI stack** — with **auditability**, **security**, and **trust** built in. Rather than relying on opaque cloud APIs and external data flows, **Sovereign AI** ensures that your intellectual property, business data, and AI-driven applications stay under your governance — and run on infrastructure you control. ``---`` Why it matters -------------- AI is transforming how organizations interact with data and deliver new experiences. But with AI adoption comes real risks: - Many public AI APIs are **black boxes** — you lose control over how your data is used. - AI embeddings and vectors often become **orphaned** in third-party systems. - Sensitive data may leave compliance boundaries — creating **regulatory risk**. - AI results may be **non-auditable**, raising concerns in regulated industries. - Application **traceability** and **explainability** are often missing. **Sovereign AI** answers these challenges: - **Own your data and embeddings** — stored in Postgres and your object storage. - **Own your models and AI stack** — served through Pipelines and Model Serving. - **Own your Knowledge Bases** — fully governable, versioned, and auditable. - **Own your AI workloads** — running in Kubernetes environments you control. - **Own your Assistant behavior** — defined, governed, and observable in Agent Studio. ``---`` What is Sovereign AI? --------------------- **Sovereign AI** is an architecture and operating model where: - **Data stays inside your trust boundaries**. - **AI pipelines are fully transparent and auditable**. - **Model serving happens on infrastructure you manage**. - **Embeddings are stored where you control them** (Postgres, object storage). - **Assistant behavior is governed and versioned**. - **APIs exposed to your users are secured and observable**. - You can **choose whether and how external APIs are involved** — no hidden dependencies. **Sovereign AI is a spectrum** — not an all-or-nothing choice: - Some organizations will want **fully self-managed, zero external API** architectures. - Others will use hybrid models — **private Knowledge Bases + public LLMs**, carefully governed. - EDB PG AI supports this full spectrum — and ensures you are never locked into an external stack. ``---`` How EDB PG AI enables Sovereign AI ---------------------------------- EDB PG AI is designed from the ground up for Sovereign AI: - **Data-first architecture** — Knowledge starts in your databases and object storage. - **Vector Engine** — Your embeddings live in Postgres or controlled storage. - **Pipelines** — You control every step of the data-to-embedding-to-query pipeline. - **Model Serving** — Models run on your Kubernetes infrastructure, under your governance. - **Agent Studio** — Rulesets, Tools, Structures, and Assistants are all governed assets. - **Observability** — Full audit trails and monitoring of model and Assistant activity. Key components: ^^^^^^^^^^^^^^^ - :ref:`Vector Engine (PGvector) ` — Vector search in your Postgres clusters. - :ref:`Pipelines and data preparation ` — Controlled creation of Knowledge Bases and embeddings. - :ref:`Model Serving Reference Manual ` — Kubernetes-native, GPU-accelerated model inference. - :ref:`Gen AI in Hybrid Manager ` — Governed Assistant and Tool development. You can also selectively integrate **external LLM APIs** — but **with full visibility and choice**. ``---`` Why Hybrid Manager? ------------------- Today, EDB PG AI’s **strongest Sovereign AI model** is delivered with **Hybrid Manager™** (EDB): :ref:`See: Hybrid Manager Gen AI Workloads ` Hybrid Manager provides the architecture to: - Deploy Pipelines and Model Serving on **EDB-optimized Kubernetes**. - Manage Knowledge Bases with **full lifecycle control**. - Monitor and trace all AI activity — at the database, Pipeline, and Assistant levels. - Govern **model deployment**, image provenance, and retriever configurations. - Enable **secure API access** to Assistants and semantic search. - Implement fine-grained **data residency and trust boundaries**. **Full stack control** — from your Postgres database → embeddings → Pipelines → model inference → Assistants and APIs — all observable and auditable. ``---`` Summary ------- - **Sovereign AI** is about trust, transparency, and control. - EDB PG AI gives you a complete stack to implement Sovereign AI — with no lock-in. - You choose how **open or closed** your AI architecture is — but you never lose **control of your data**. - Hybrid Manager provides a production-proven platform for deploying **trusted, governed AI** in the enterprise. ``---`` Learn more ---------- - :ref:`AI Factory Overview ` - :ref:`Vector Engine (PGvector) ` - :ref:`Pipelines and data preparation ` - :ref:`Model Serving Reference Manual ` - :ref:`Gen AI in Hybrid Manager ` ``---``