Shaped
Real-time retrieval infrastructure for personalized search, recommendations, and agent context, with ranking models and a feedback loop.
Shaped has been acquired by Whatnot.
About Shaped
Shaped is a real-time retrieval engine and vector database for personalized search, recommendations, and agent retrieval. It connects to existing data sources, transforms and enriches data, and defines retrieval engines, ranking models, and pipelines through YAML. A query can combine semantic and lexical retrieval, business-rule filters, learned scoring, and diversity or exploration controls. Results can incorporate user or item context and improve from interaction feedback such as clicks, likes, and skips. Developers can query Shaped through a REST API, Python SDK, TypeScript SDK, MCP, or ShapedQL. It suits e-commerce catalogs, marketplaces, media libraries, and agent knowledge retrieval where fresh data and personalized ranking matter. Shaped handles embedding generation, model training, indexing, deployment, and automatic scaling, reducing the number of separate retrieval and ranking systems a team must operate. Adoption still requires connecting production data, defining schemas and pipelines, and evaluating ranking behavior against business goals. Some connectors, including Amplitude, require the Standard Plan or higher. The homepage advertises $100 in free credits with no credit card required, allowing teams to test the service before broader production use.
Best for
Best use cases
Combine semantic and lexical search with learned reranking
React to clicks, likes, skips, and changing catalog data in real time
Apply business rules, diversity controls, and value functions during retrieval
Compare before choosing
Consolidates retrieval, ranking, training, and feedback into one managed system
Adds user-aware ranking beyond a conventional vector store
Requires more data and ranking setup than a basic document similarity API
Features
Personalized search
Blend semantic and keyword retrieval, then rerank results with user context, learned models, business rules, and freshness signals.
Product and content recommendations
Use interaction feedback to update recommendations for e-commerce, marketplace, or media experiences in real time.
Agent retrieval
Retrieve relevant, filtered context for agents through ShapedQL, REST APIs, SDKs, or MCP while keeping indexes current.
Pros
- Combines retrieval, ranking, training, and interaction feedback in one managed workflow.
- Supports semantic, lexical, and personalized search with filters and business rules.
- Offers REST, Python, TypeScript, MCP, and ShapedQL integration options.
Cons
- Production adoption requires data connectors, schemas, pipeline configuration, and ranking evaluation.
- The Amplitude connector requires the Standard Plan or higher, adding a plan constraint for that workflow.
- The homepage states only the free-credit amount, so teams need to contact Shaped for broader plan terms.
Pricing Plans
Free credits
- No credit card required
- Access for testing Shaped retrieval workflows
- Connect data and evaluate search or recommendation use cases
Standard Plan or higher
- Required for the Amplitude connector
- Supports real-time Amplitude event streaming through AWS Kinesis
- Includes Shaped table provisioning through CLI or API
* Pricing information is for reference only. Please refer to the latest announcement on Shaped's official website.