Shaped

Shaped is a real-time context engine specifically designed for agentic AI systems, functioning as a vector database with an integrated feedback loop. Its core capability is to reduce the computational cost and noise in AI agent operations by delivering only the most relevant contextual results—typically 10 precise answers instead of 200 noisy ones. This drastically cuts per-answer costs from approximately $1.50 for a do-it-yourself approach down to $0.03 using Shaped. The platform enables developers to connect their own data sources, train custom models, and perform queries across text, user identifiers, or session contexts, all with sub-50 millisecond latency. The engine supports a hybrid search approach that combines semantic search, keyword search, and vector similarity to retrieve relevant results from multiple indexes in a single query. This is achieved through ShapedQL, a dedicated query language that allows users to specify retrieval by text, user ID, or item ID. The system also incorporates hard constraints and business rules, enabling precise filtering based on real-world requirements like access permissions, content freshness, or geographic restrictions. Additionally, it supports scoring via machine learning models and value functions, allowing results to be ranked according to learned user preferences or business objectives. A reordering layer introduces diversity and exploration, ensuring that results are not only accurate but also varied, which is essential for recommendation systems or exploratory search. The feedback loop is a distinguishing feature: as users interact with the retrieved results, those interactions are fed back into the system to fine-tune the ranking and scoring models continuously. This means the database gets smarter over time without m