Shaped

Shaped is a relevance engine for search, recommendations and agent memory, described by the vendor as a vector database with a feedback loop. Customers connect their data and Shaped handles embeddings, models and data freshness, returning results in milliseconds. Unlike a plain vector store, it knows who is asking, so results can be personalized per user and improve with every interaction. Retrieval, filtering, scoring and reordering are combined in a single query written in ShapedQL, a SQL style interface. It supports text, image, lexical, semantic and hybrid search, and can retrieve by text, user ID or item ID. Business rules such as in stock status, margin and merchandising constraints can be applied, and diversity or exploration can be added to the ordering. The platform offers more than 30 native connectors and is accessible through Python and TypeScript SDKs or MCP. Typical uses are ecommerce search, personalized product recommendations across home, product, cart and post purchase surfaces, and context retrieval for AI agents. It is aimed at engineering teams who want to replace a stack of separate retrieval, ranking and pipeline services. The home page offers $100 in free credits with no credit card required. No plan prices are published, and there is no pricing page. The site states that Whatnot has acquired Shaped.