Vespa
What is Vespa?
Vespa is an open-source engine for real-time serving, search, recommendation, and ranking of large-scale data and machine-learned models (search and data serving platform).
- Real-time search and retrieval over structured and unstructured data (search engine).
- Low-latency serving and inference for Machine Learning (ML) models at scale (ML serving).
- Support for vector search, semantic search, and hybrid retrieval combining text, vectors, and structured filters (information retrieval).
- Distributed, horizontally scalable storage and query execution with automatic sharding and replication (distributed data platform).
- Integrated ranking framework with feature evaluation, ranking expressions, and A/B testing support (relevance and ranking platform).
Show more
More About Vespa
Vespa is an open-source engine for large-scale data serving and search, designed to serve feature-rich search, recommendation, and personalization applications in real time (search and data serving platform). It addresses workloads where applications must evaluate complex relevance logic, query large document or item corpora with low latency, and integrate machine-learned models directly into ranking and decision logic. Vespa is used for applications such as product search, content discovery, recommendation systems, and personalized feeds in environments where data volume, query rate, and relevance complexity are high.
At its core, Vespa provides distributed indexing and querying for both structured and unstructured data (search engine). It supports full-text search, filters, aggregations, and facets alongside support for tensors and vector representations used in modern information retrieval (vector search platform). Vespa enables hybrid retrieval that combines lexical matching, dense vector similarity, and structured constraints within a unified query model, allowing enterprises to implement semantic search, neural retrieval, and business-rule-aware ranking in one system.
Vespa includes a model-serving and ranking framework that lets operators deploy and execute ML models directly inside the serving layer (ML serving). Models can be used for ranking, scoring, and other decision logic as part of query evaluation. Vespa supports ranking expressions and feature evaluation, where features can be computed from document fields, query parameters, tensors, and external signals (relevance and ranking). This enables experimentation with multiple ranking functions, A/B testing, and online tuning of relevance without redesigning the underlying data infrastructure.
The system is built as a distributed platform with automatic sharding, replication, and cluster management across content nodes and stateless container nodes (distributed systems). Vespa is designed to handle high query throughput and continuous updates, providing near real-time visibility of new or changed documents (real-time indexing). It integrates storage and serving, so applications can query and aggregate over large data sets without an external data warehouse for serving use cases.
From an architecture perspective, Vespa exposes HTTP/JSON-based APIs for feeding data, issuing queries, and managing configuration (API platform). It supports configuration and application packaging through an application model that defines document schemas, search definitions, ranking profiles, and services (configuration management). Vespa can be deployed on-premises (on-prem) or in cloud environments and offers tooling and documentation for cluster operations, metrics, and monitoring (operations and observability).
In enterprise environments, Vespa functions as an online serving layer that combines search, recommendation, and model inference, reducing the need to stitch together separate search engines, feature stores, and model servers. Its ability to evaluate complex ranking logic and ML models at query time positions it in categories such as search infrastructure, online feature and model serving, and personalized content delivery platforms.