Vespa.ai

Vespa.ai is a company developing a platform for low-latency computations over large, evolving data sets. A part of Yahoo between 2003 and 2023, Vespa.ai was spun out of the company to offer services to a larger number of customers. Vespa allows organizations to search, make inferences in, and organize vectors, tensors, text, and structured data, at serving time and any scale. It supports vector search (ANN - Approximate Nearest Neighbour), lexical search, and search in structured data, from the same query. Vespa.ai open sources its technology under an Apache 2.0 license.

Vespa stores and indexes structured, text, and vector data, enabling faster queries, selection, processing, and machine-learned model inference over the data at serving time at any scale. Functionality can be customized and extended with application components hosted within Vespa. The platform aims to allow developers to build applications that scale to large amounts of data and high loads without sacrificing latency or reliability. A Vespa application consists of a number of stateless Java container clusters and zero or more content clusters storing data.

Architecture of the Vespa platform.

Stateless container clusters host components that process incoming data and/or queries and their responses. These components provide functionality belonging to the platform, like indexing transformations and the global stages of query execution, but can also be the middleware logic of the application. Application developers can configure their Vespa system with a single stateless cluster that performs all such functions, or create different clusters for each kind of task. The container clusters then pass queries and data operations on to the appropriate nodes in content clusters.

Content clusters in Vespa are responsible for storing data and executing queries and inferences over the data. Queries can range from simple data lookups for content serving to complex conditions for selecting the relevant data, ranking it using machine-learned models, and grouping and aggregating the data across all nodes participating in the query. Operations provided by Vespa scale to more content, more expensive inference, and higher query volume simply by adding more nodes to the content clusters.

Use cases of the Vespa platform include the following:

• Search
• Recommendations & personalization
• Conversational AI
• Semi-structured navigation
• Personal search
• Typeahead suggestions
• Question answering

Vespa began as a search engine in 1997, becoming part of Yahoo through the Overture deal in 2003. Vespa went on to be used across Yahoo in search, recommendation, and ad serving. Vespa open-sourced in 2017 and began offering services to external customers in 2021, including Spotify, Wix, major financial institutions, and others. On October 3, 2023, it was announced that Vespa.ai would be spun out of Yahoo, becoming a separate company. Yahoo owns a stake in the new entity, holds a seat on its board of directors, and remains one of its largest customers. At the time of the new company's founding, Vespa technology powered approximately 150 of Yahoo's applications, serving personalized content across all of Yahoo's pages in real time. Collectively, these applications serve nearly one billion individuals, processing 800,000 queries a second.

Creating a separate company aims to bring Vespa's technology to a larger number of customers and offer existing customers access to its cloud service. The new company is based in Trondheim, Norway, led by the CEO Jon Bratseth (one of the four founders) alongside Kim O. Johansen (COO), Frode Lundgren (CTO), and Kristian Aune (Head of Customer Success).