MongoDB has recently announced the public preview of its Embedding and Reranking API on MongoDB Atlas. The new API gives developers direct access to Voyage AI’s search models within the managed cloud database, enabling them to create features such as semantic search and AI-powered assistants within a single integrated environment, with consolidated monitoring and billing.
This option consolidates the components needed to build AI retrieval on a single platform. According to MongoDB, this new API is database-agnostic, and it can be integrated into any tech stack or database. It is designed for teams building retrieval-powered AI systems, from semantic search and RAG to AI agents. Thibaut Gourdel, senior technical product marketing manager at MongoDB, and Wen Phan, staff product manager at MongoDB, write:
Building AI retrieval today means stitching together databases, vector search, and retrieval model providers — each introducing operational complexity. To address this, we’re introducing the Embedding and Reranking API on MongoDB Atlas.
Source: MongoDB blog
One of the main announcements at the .local San Francisco event, the Voyage 4 series is now available and consists of four different models: voyage-4-large, voyage-4, voyage-4-lite, and the open-weights voyage-4-nano. While previous generations of embedding models required using identical models to embed both queries and documents, Voyage 4 provides text embedding models that work in the same embedding space, so teams can, for example, store data using voyage-4-large and run queries with any Voyage 4 model.
Furthermore, automated embedding in vector search is available in preview in the community edition, and Lexical Prefilters for MongoDB Vector Search is in public preview, providing developers with text and geo analysis filters alongside vector search. Deepak Goyal comments on LinkedIn:
I spent 3 hours yesterday debugging a 12-hour sync lag in our vector store. It’s a “Sync Tax” that almost every AI team is paying right now. (…) If your data is 24 hours old, your RAG isn’t “intelligent”—it’s just a well-indexed archive (…) By unifying the flow, we’re seeing a shift (…) Specialized vector stores are starting to feel like the external GPUs of the AI world—powerful, but for 90% of production use cases, “integrated” is winning on speed and simplicity.
The embedding models support dimensions from 256 to 2048 and quantization, allowing developers to balance accuracy, cost, and speed. In addition to general models, Voyage provides options designed for specific fields, whole-document analysis, multimodal data, and reranking in multi-step search systems. Gourdel and Phan emphasize that while MongoDB Atlas already supports built-in vector search capabilities, the new API brings simplicity:
This matters for building production AI systems. Scaling LLM applications requires delivering the right context at the right time, which means tightly integrating operational data with high-performance search.
The integration of Voyage AI’s capabilities into MongoDB Atlas has been expected and questioned by the community since MongoDB announced the acquisition of Voyage AI almost a year ago.
The Embedding and Reranking API is currently in preview. The “Voyage AI Quick Start” tutorial is available as a Python notebook on GitHub.
