Highlights:

  • Voyage AI has developed more than six AI models for generating embeddings, with its flagship model, voyage-3-large, launching last month.
  • MongoDB plans to integrate Voyage AI’s models into MongoDB Atlas later this year.

MongoDB Inc. acquired Voyage AI Inc., a startup specializing in artificial intelligence models for generating embeddings.

The financial terms of the deal were not disclosed. Voyage AI had previously secured USD 28 million in funding from investors, including Snowflake Inc. and Databricks Inc.

MongoDB, a publicly traded company on Nasdaq, is known for its popular document database, which differs from traditional relational databases by allowing records to be stored without a fixed format. This flexibility simplifies application development.

The company offers its database in multiple editions, including MongoDB Atlas, a managed cloud-based version that eliminates infrastructure management and streamlines tasks such as patching and query optimization.

Atlas is designed to support various use cases, including AI applications. The acquisition of Voyage AI aims to further enhance these capabilities.

AI models do not process business files in their original format; instead, they first convert them into embeddings. These mathematical structures capture key details about a file and define its relationship to other records in a database. Typically, embeddings are created using AI models.

Voyage AI has developed more than six AI models for generating embeddings, with its flagship model, voyage-3-large, launching last month. This model claims to surpass competing embeddings from OpenAI and Cohere Inc. by 9.7% and 20.7%, respectively, in terms of quality.

Additionally, the company offers specialized models tailored for specific industries. These models are optimized to generate embeddings from code files, legal documents, and financial data.

Voyage AI provides its embedding models alongside a suite of AI-powered rerankers. These models refine search results by prioritizing the most relevant items. When an AI application retrieves a dataset to answer a query, a reranker helps identify and display the most relevant data points first.

MongoDB plans to integrate Voyage AI’s models into MongoDB Atlas later this year. This addition will allow developers to efficiently convert records into embeddings and store them directly in the database. AI applications built on Atlas will also benefit from Voyage AI’s rerankers, enabling them to identify and retrieve the most relevant data for a given task.

The company views this integration as a way to streamline application development. Typically, software teams manage standard data and AI-generated embeddings in separate databases. By incorporating Voyage AI’s technology, MongoDB aims to make it easier to store both data types within Atlas, simplifying the development process.

“Instead of implementing workarounds or managing separate systems, developers can generate high-quality embeddings from real-time operational data, store vectors, perform semantic search, and refine results — all within MongoDB,” MongoDB Chief Executive Officer Dev Ittycheria reported.

Following the integration of Voyage AI’s models into Atlas, MongoDB plans to enhance their capabilities. The company will expand support to include multimodal data, such as images and videos. It also aims to introduce industry-specific features tailored for sectors like legal and finance.