MongoDB.native San Francisco 2026: Ship Manufacturing AI, Sooner

As we speak at MongoDB.native San Francisco, we introduced capabilities that collapse the space between AI prototype and manufacturing.

Constructing AI purposes means fixing actual issues: protecting conversational context clear and queryable, retrieving the correct data from hundreds of previous interactions, connecting AI brokers to your information with out customized plumbing. These aren’t theoretical challenges, they’re the friction factors that sluggish groups down on daily basis.

The AI period calls for extra out of your information platform. MongoDB provides you every little thing you should construct shortly.

Voyage AI: the most effective will get higher

Embedding fashions could make or break AI search experiences. We’re proud that voyage-3-large has been the world’s top-performing embedding mannequin on Hugging Face’s RTEB benchmark since its inception. 

However we didn’t relaxation on our laurels. There’s a brand new mannequin on the prime of the charts.

As we speak, we’re happy to announce that the Voyage 4 mannequin household is now usually obtainable. The very best simply bought higher. The voyage-4 collection fashions function in a shared embedding house, permitting for cross-model compatibility and unprecedented flexibility to optimize for accuracy, velocity, or value. This launch additionally consists of voyage-4-nano, our first open-weight mannequin obtainable on HuggingFace, excellent for native growth.

Moreover, we’re launching the brand new voyage-multimodal-3.5 mannequin, which has been particularly educated to assist video content material alongside textual content and pictures. For builders constructing multimodal AI purposes, this represents a big leap ahead in dealing with numerous content material sorts inside a single retrieval system. Better of all, upgrading is remarkably easy—you possibly can merely change the mannequin parameter to “voyage-multimodal-3.5” in your API name, immediately unlocking video capabilities without having to refactor your present codebase or change your utility structure.

Lastly, we’re saying the public preview of the Embedding and Reranking API on MongoDB Atlas, offering API assist for Voyage AI fashions. Whereas enabling standalone utilization of the fashions with any know-how stack, the API advantages from the strong safety and scalability requirements of MongoDB. By bringing crucial parts right into a single management airplane and interface, it eliminates the necessity to handle separate distributors and considerably reduces operational overhead.

Automated Embedding, comfort constructed into MongoDB Group

Persistence issues. An AI with amnesia isn’t useful; customers want methods to recollect context from minutes, hours, and weeks in the past. Each interplay is a goldmine of preferences, patterns, and conduct that ought to make the subsequent interplay smarter.

However storing dialog historical past in a database is not sufficient. Easy storage solves nothing if you cannot retrieve the correct data on the proper time. The actual problem is clever retrieval: discovering related context throughout hundreds of previous interactions, filtered by metadata and consumer attributes, with out your system buckling below manufacturing load. That is the place vector search turns into crucial—enabling semantic search that captures which means, not simply key phrases, whereas working in your real-time operational information. And that is the place MongoDB’s method eliminates a significant ache level: the necessity to sync information between separate methods for vectors and utility information.

Till now, producing and storing these vectors required overhead—growth time, infrastructure administration, and cognitive load. Not.

We’re introducing Automated Embedding for MongoDB Group Version in public preview. MongoDB Group Version now handles the complexity of managing embedding fashions routinely, giving builders high-accuracy semantic search within the database whereas sustaining flexibility to make use of any LLM supplier or orchestration framework. Automated Embedding affords one-click automated embedding straight inside MongoDB, which eliminates the necessity to sync information and handle exterior fashions. It’s a simple solution to get prime quality embedding natively.

Greatest-in-class retrieval should not require infrastructure work—Automated Embedding in MongoDB Vector Search delivers on that promise. Automated Embedding in MongoDB Vector Search is offered now in Group Version, with Atlas entry coming quickly.

Exact textual content filtering for superior search use circumstances

As we speak, we introduced the launch of Lexical Prefilters for Vector Search. This addresses a long-standing request from builders constructing semantic search interfaces who want superior textual content filtering alongside vector operations.

The brand new syntax permits highly effective textual content filtering capabilities—fuzzy matching, phrase search, wildcards, and geospatial filtering—as prefilters for vector search. This leverages full textual content evaluation capabilities whereas sustaining the semantic energy of vector search. We have launched a brand new vector information sort in $search index definitions and a vectorSearch operator inside the $search aggregation stage to make this work seamlessly.

This replaces the knnBeta operator with a cleaner, extra highly effective method. For groups already utilizing lexical and vector search collectively, this supplies a simplified migration path with considerably expanded capabilities.

Clever help wherever you’re employed

MongoDB’s clever assistant is usually obtainable in MongoDB Compass. The assistant supplies in-app steerage for debugging connection errors, optimizing question efficiency, and studying finest practices, all with out leaving your growth surroundings. You possibly can even question your database utilizing pure language by read-only database instruments that require your approval earlier than execution, permitting for deeper contextual consciousness of your information.

The assistant was constructed to handle actual friction: builders switching between a number of instruments and documentation tabs, ready for assist responses, or getting generic recommendation from general-purpose AI chatbots that do not perceive MongoDB-specific contexts. Now, tailor-made steerage is offered immediately, proper the place you are working.

The modernized Atlas Knowledge Explorer interface brings the Compass expertise straight into the Atlas net UI, addressing a crucial hole for groups with safety insurance policies that prohibit desktop utility utilization. Customers can now carry out refined question growth, optimization, bulk operations, and complicated aggregations—all with AI help—throughout all MongoDB Atlas clusters in a unified net interface.

Whether or not you are troubleshooting a connection challenge, optimizing a sluggish question, or studying the best way to construction an aggregation pipeline, the clever assistant delivers MongoDB-specific experience with out context switching. Strive the clever assistant within the modernized Atlas Knowledge Explorer now.

The engine behind MongoDB Search and Vector Search is now obtainable below SSPL

Lastly, mongot, the engine powering MongoDB Search and Vector Search, is now publicly obtainable below SSPL. Whereas nonetheless in preview, after years of growth and funding, we’re making the supply code of this core know-how obtainable to the group, increasing our unified search structure past Atlas to each MongoDB deployment. 

mongot runs individually from mongod, MongoDB’s core database course of, and is the inspiration that makes highly effective search native to MongoDB. Releasing mongot below SSPL means full transparency for safety audits and debugging advanced edge circumstances. Builders can dive into mongot‘s structure, perceive how search and vector operations work below the hood, and assist form the way forward for search at MongoDB.

A contemporary information platform that evolves along with your wants

These bulletins replicate our dedication to anticipating what builders want as AI growth matures. Vector search, time collection, stream processing, queryable encryption, Atlas itself—we have constantly delivered on rising necessities. “If you happen to’re constructing an early-stage firm that’s going to scale very quickly, you want a database answer that is not going to interrupt below the load of an enormous quantity of customers,” stated Eno Reyes, Co-founder and CTO of Manufacturing unit. “You want a fast-moving workforce with a dependable answer, and there actually is one choice on this house—and it is MongoDB.”

Rabi Shanker Guha, CEO of Thesys, put it this manner: “MongoDB helps us transfer quick in an ever-changing world. The very best database is the one you don’t have to consider—it simply works precisely the place and the way you want it. That’s MongoDB for us.”

Ship quicker, scale confidently

Every functionality we introduced immediately addresses actual friction within the AI growth workflow and within the developer expertise. We’re not asking builders to decide on between structured information and vectors, between efficiency and adaptability, or between speedy iteration and manufacturing readiness.

The promise is easy: ship quicker, scale confidently, and give attention to what makes your AI utility distinctive—not on managing database infrastructure. In an ecosystem crowded with level options and retrofitted legacy methods, MongoDB is a contemporary information platform constructed for the lengthy haul.