MongoDB Atlas precision accuracy changes AI context retrieval
Retrieval accuracy can shape both what an AI application decides and what it costs to run. At MongoDB.local Build Fest, MongoDB announced new capabilities aimed at bringing high-precision retrieval into its intelligent data platform.
At the centre of the announcement is MongoDB Atlas precision accuracy, with new retrieval capabilities built around Voyage AI models. The company said its Voyage AI embedding models hold the top spot on the Retrieval Embedding Benchmark (RTEB).
MongoDB Atlas retrieval moves closer to live data
For developers, one of the problems highlighted by MongoDB is the complexity of keeping operational data, embedding pipelines and vector stores in sync.
Automated Embedding in MongoDB Atlas, powered by Voyage AI models, is designed to remove much of that work. Atlas automatically embeds new documents as they are written and re-embeds existing ones when they change.
That creates a more direct path between operational data and AI context retrieval. Instead of relying on a separate copy that can become outdated, agents can retrieve context from live data.
The Financial Times is cited as one example. After consolidating search on the MongoDB platform, it said Automated Embeddings helped improve AI-powered semantic search accuracy while keeping retrieval costs in check across more than 100,000 searches a day.
Atlas Embedding API adds another route
MongoDB is also introducing the Atlas Embedding and Reranking API. It gives applications, including those running outside MongoDB, access to the company’s embedding and reranking models through a single endpoint.
For teams building retrieval-based applications, this changes where the retrieval layer sits. The API can provide access to the models without requiring the application itself to manage the underlying embedding and reranking setup.
Eve, a legal AI platform, is using the API to surface relevant material across the life of a case. Its focus is retrieval quality and simplifying the infrastructure behind its RAG layer.
Automated embeddings MongoDB brings context up to date
The announcement also targets coding agents. MongoDB introduced voyage-code-4, a model designed specifically for agentic code retrieval. The company said the model provides higher precision and lower cost than previous models for this use case.
Another capability extends retrieval to streaming data through Vector Search in Atlas Stream Processing. The aim is to give agents working with live events access to the same retrieval accuracy available for data at rest.
Read More:
ZOTAC Strengthens Channel Ecosystem With Direct App
NativeDefence Unveils NativeX: One Platform for the Modern Security Operations Centre
2 Min read