Shaped

Shaped is a real-time retrieval engine designed specifically for developers who need to build highly relevant search, personalized feeds, and AI agent memory into their applications. It works by letting you connect your data sources—like data warehouses, analytics apps, and catalogs—and then query them using a custom SQL-like language called ShapedQL. This allows you to build complex, multi-stage ranking pipelines that combine hybrid search, business rule filtering, and machine learning models to return the most relevant results in under 50 milliseconds.

What makes Shaped stand out is its three-layer architecture and its focus on treating user context as a first-class input. Instead of just retrieving flat documents, it uses a unified data layer with over 30 connectors, an intelligence layer that continuously trains ML models on user-item interactions, and a query layer that compiles your SQL into optimized ranking pipelines. This means you can personalize results based on who a user is, not just what they typed. Its declarative SQL approach also lets teams experiment and deploy new ranking logic in minutes instead of months.

This tool is a major benefit for product and engineering teams at companies where user engagement directly impacts revenue, like e-commerce platforms, social media apps, and content streaming services. Real-world use cases include building a 'For You' video feed that avoids previously watched content, creating a product search that boosts items with high conversion rates, or powering an AI agent's memory retrieval with rich, contextual understanding. It’s built for teams that have outgrown simpler search solutions and need fine-grained control over relevance without building the entire infrastructure in-house.