Affan Dar, Vice President of Engineering, PostgreSQL at Microsoft
Charles Feddersen, Partner Director of Program Management, PostgreSQL at Microsoft
Today at Microsoft Build, we’re pleased to announce the public preview of Azure HorizonDB, a new enterprise-ready Postgres-compatible database service designed to meet the needs of modern AI applications, alongside a set of enhancements to our PostgreSQL tooling in Visual Studio Code to further streamline the developer experience.
Postgres is rapidly solidifying its role as a foundational layer in modern data architectures, with accelerating adoption across industries. For developers, it has become the preferred platform for new application development, driven by its extensible architecture, mature extension ecosystem, and adherence to open standards and APIs. At the same time, enterprises are choosing Postgres to re-platform and modernize existing systems, taking advantage of its ability to support a broad range of operational workloads while enabling advanced capabilities such as vector-based data access all within a single, interoperable platform.
A Postgres Platform Grounded in Security, Resilience, Scale, and Performance
Azure HorizonDB is purpose-built to meet these demands, combining the flexibility developers expect from Postgres with the operational rigor enterprises require. It extends the core Postgres engine with cloud-native capabilities such as integrated identity, fine-grained network and security controls, and seamless lifecycle management, while preserving full compatibility with the open ecosystem of extensions and tools. At the same time, HorizonDB introduces advanced, natively integrated capabilities like vector data support and AI model management, enabling new classes of intelligent applications without sacrificing transactional integrity or developer productivity.
These capabilities are backed by a platform designed for enterprise performance and scale. HorizonDB supports databases up to 128 TB, scales out with up to 15 read replicas for high-throughput workloads, and delivers sub-millisecond commit latency across availability zones for low-latency transactions and high availability. This combination is critical for modern applications that require consistent performance under load, including high-concurrency transactional systems, real-time AI-driven interactions, and globally distributed services.
The result is a unified platform that scales from the first line of code to globally distributed, mission-critical systems. Enterprise adoption ultimately depends on trust in the platform itself. Azure HorizonDB delivers this with native integration into Microsoft Entra ID for centralized identity and access control, private endpoints for network isolation, and built-in encryption to protect data at rest and in transit. These capabilities are essential for meeting compliance requirements and enabling organizations to run mission-critical workloads with confidence, without added complexity.
This foundation is critical for any application, but it becomes indispensable for AI, where secure access to data and controlled model interaction underpin every intelligent experience. Building on this, HorizonDB introduces a set of integrated AI capabilities designed to bring intelligence directly into the database.
Run Fast, Memory-Efficient Vector Search with DiskANN
HorizonDB brings high-performance vector search directly into Postgres through DiskANN with spherical quantization. This enables efficient, low-latency similarity search at scale while significantly reducing memory and storage overhead. Spherical quantization works by normalizing vectors and encoding them into compact representations that preserve angular distance, allowing the system to compare vectors efficiently with minimal loss in accuracy. The result is the ability to index and query large embedding datasets within the transactional engine itself, making vector search a first-class capability rather than an external dependency.
"HorizonDB is compelling because it brings a PostgreSQL-compatible foundation, AI-native capabilities and enterprise-grade controls closer to the operational data layer."
Jennings Balavari, Founder, Opsen AI
Build Smarter Apps with Hybrid Search in Postgres
HorizonDB supports hybrid search by combining vector similarity through pgvector with full-text search enabled via the pg_textsearch extension, allowing applications to match both semantic meaning and precise keyword relevance in a single query. This enables more accurate, context-aware results, such as blending intent-driven retrieval with exact term matching for search, recommendations, or RAG scenarios. By unifying these capabilities within Postgres, HorizonDB improves result quality while simplifying application design without the need for external search systems.
Operationalize AI with Built-In AI Model Management
Working with vectors requires models to generate, interpret, and evolve embeddings, making model lifecycle a core part of the application stack. HorizonDB introduces integrated AI model management to simplify how models are registered, versioned, and governed alongside data, including built-in support for generative GPT models and ranking models. For example, GPT models can be used to generate summaries, responses, or structured outputs directly from application data, while ranking models enable relevance scoring for search results or recommendations over vector results. By managing these models alongside the data they operate on, HorizonDB ensures consistency, traceability, and control, creating a unified environment where models and data evolve together.
“As we build a multi-tenant, AI-driven commerce platform, HorizonDB has been particularly compelling in two areas: scale and how close AI capabilities are to the data itself. Running vector search, filtering, and model-driven workflows directly inside the database removes a lot of the complexity we’d normally manage across separate services."
James Frawley, CIAO, ReFiBuy
Bring AI into SQL with AI Functions
With models managed in place, AI Functions provide a direct way to invoke them from within SQL and application logic. These functions are implemented through the azure_ai extension, which brings model invocation directly into the Postgres engine. This allows developers to embed inference into queries and transactions, eliminating the need for external orchestration. By bringing model execution closer to the data, AI Functions reduce latency, simplify application design, and make intelligent behavior a natural extension of existing Postgres workloads.
"What stood out with HorizonDB is that it aligns closely with how we already think about the problem. Instead of stitching together multiple components, it brings transactional data, vector search, and AI capabilities into a single platform, which simplifies the architecture without forcing a complete rethink."
Mohsin Shafqat, Director Software Engineering, Nasdaq
Run Reliable, Event-Driven Workflows with AI Pipelines
Finally, AI Pipelines operationalize these capabilities through reliable, event-driven workflows for model execution and data processing. Pipelines execute on data changes, enabling real-time asynchronous reactions without external orchestration and ensuring consistent, repeatable behavior as data evolves. Combined with model management and AI Functions, they turn embedded intelligence into something that can be run, scaled, and trusted in production, while inheriting the database’s high availability and failover characteristics for resilience. Pipelines can also be visualized and observed in real time through the Visual Studio Code extension for PostgreSQL, giving developers and operators immediate visibility into execution flow, state, and outcomes
Modern Unified Experience for Data, AI, and Operations in VS Code
As intelligence becomes a core part of the data platform, the developer and operator experience becomes equally critical. HorizonDB extends seamlessly into Visual Studio Code with enhanced PostgreSQL tooling that works across any Postgres deployment, not just HorizonDB. Features like AI-assisted query plans and integrated monitoring enable faster debugging and optimization, helping teams understand both database performance and AI-driven behaviors.
At the same time, for Azure-based deployments, the experience is deeply integrated with platform capabilities, enabling management of networking configuration, server parameters, and server logs directly from the development environment, streamlining operations across application and infrastructure layers.
Azure HorizonDB brings together enterprise-grade security, deep Postgres compatibility, and a modern AI-native data platform, all engineered for developers. It scales efficiently across workloads, from transactional systems to intelligent applications, while delivering a world-class, Azure-integrated experience in Visual Studio Code for both developers and operators.
Ready to get started with Azure HorizonDB?
Azure HorizonDB is now available in public preview in Australia East, Central US, Sweden Central, West US 2, and West US 3 regions. Additionally, East US, Canada Central, Indonesia Central, Italy North, Japan East, Korea Central, and Poland Central will be available in the coming weeks.
You can get started today by creating a new HorizonDB instance using the Azure portal, API’s, or the Visual Studio Code extension for PostgreSQL to begin exploring these capabilities firsthand.
To learn more, dive deeper into our documentation and sign-up today to try AI model management in a limited preview.