# Vespa.ai Vespa is an AI search platform for building search, retrieval-augmented generation, recommendation, personalization, and agent applications over text, vectors, tensors, and structured data. It is designed for developer teams that need configurable ranking and distributed operation at production scale. - Findings24: - Site: - Preços: Freemium - Categorias: Motor de Busca IA, API de pesquisa com IA, Banco de dados vetorial ## O que é Vespa.ai? Vespa is an AI search platform for applications that query, organize, rank, and make inferences over text, vectors, tensors, and structured data. It combines lexical, vector, and structured retrieval with configurable machine-learned ranking, allowing teams to build search and retrieval workflows without separating the search engine from the ranking and inference layer. The platform supports search and hybrid retrieval, RAG, recommendation and personalization, semi-structured navigation, and personal or private search. Its distributed architecture is intended for workloads with large and frequently changing datasets; the product site describes operation across billions of items, thousands of queries per second, and latencies below 100 milliseconds, depending on the application and configuration. Vespa can be used through Vespa Cloud, including deployments on Vespa-managed AWS and Google Cloud accounts, or in customer-owned dedicated accounts where the data plane and data remain within the customer’s environment. Vespa Cloud also provides managed operations, automated scaling, continuous deployment, and upgrades. ## O que Vespa.ai pode fazer? ### Multi-modal retrieval Combine text search, vector and tensor search, and structured field conditions in boolean query expressions. Vector fields can contain multiple vectors, while structured data supports types including arrays, maps, and structs. ### Configurable machine-learned ranking Build rank profiles from query signals, text-match features, structured fields, geo and time data, tensors, or other inputs. Ranking functions can be authored directly or use models in ONNX and XGBoost formats. ### Multi-stage inference Apply ranking and model inference locally on content partitions, then use second-phase and global-phase reranking to spend more computation on the most promising candidates. ### Text processing and presentation Use positional text indexes, BM25 and other match features, linguistics processing such as stemming and token normalization, CJK segmentation, dynamic snippets, and word-match highlighting. ### Distributed and managed operation Scale content clusters for more data or traffic, with automatic data distribution. Vespa Cloud manages operational concerns such as failover, encryption in transit and at rest, certificates, OS patching, and platform upgrades. ## Casos de uso ### Hybrid search and RAG Retrieve passages or documents using lexical terms, embeddings, metadata, and custom ranking together, then pass the selected context to a generative AI workflow. ### Recommendations and personalization Retrieve eligible items and evaluate them with machine-learned models to select recommendations, personalized results, or ad targets under application-specific latency requirements. ### E-commerce navigation Combine text and image representations with structured filters, ranges, and other navigation conditions so users can search and refine product catalogs in one application. ### Private or personal search Search user-specific data with Vespa’s streaming search mode when indexing the full corpus, particularly vector indexes, would be inefficient for the query pattern. ## Perguntas frequentes ### What data can Vespa search? Vespa can query text, vectors, tensors, and structured data. Queries can combine operators across these field types, including lexical, exact, fuzzy, regex, range, and vector-oriented retrieval. ### How are ranking models deployed? Teams define rank profiles and can use hand-written tensor functions or machine-learned models in formats such as ONNX and XGBoost. Ranking can run in multiple stages, including local, second-phase, and global-phase evaluation. ### How is Vespa deployed? Vespa Cloud offers fully managed deployments on Vespa-managed AWS or Google Cloud accounts. The site also documents deployments in customer-owned dedicated accounts, keeping the data plane and data within those accounts. ### Is Vespa intended for frequently changing, large datasets? The platform is designed for distributed operation with automatic data distribution and cluster scaling. Its site describes support for billions of changing data items and thousands of queries per second, but actual capacity depends on the application configuration and workload. ## Informações rápidas - **Category**: AI search and retrieval platform - **Primary workloads**: Search, RAG, recommendations, personalization, and private search - **Data types**: Text, vectors, tensors, and structured data - **Deployment**: Vespa Cloud on AWS or Google Cloud; customer-owned dedicated accounts are also supported - **Managed operations**: Scaling, failover, encryption, certificates, OS patching, and platform upgrades through Vespa Cloud - **Pricing**: Usage-based Vespa Cloud pricing, with development, production, and enterprise-support tiers and $300 in free credits