MongoDB adds AI tools for production agents in Atlas

MongoDB has introduced new artificial intelligence features aimed at helping companies run AI agents in live production systems. The additions combine data retrieval, memory and infrastructure updates on its database platform.

The rollout includes automated vector embeddings in MongoDB Vector Search, a long-term memory store for LangGraph.js, performance updates in MongoDB 8.3 and cross-region connectivity support for AWS PrivateLink. The updates are intended for organisations running AI workloads across public cloud, on-premises and hybrid environments.

At the centre of the announcement is an effort to reduce the amount of separate infrastructure companies need to assemble when building AI applications. Many businesses still rely on multiple systems to manage search, data updates, memory and operational workloads, making it harder to deploy AI agents at scale.

Automated Voyage AI Embeddings in MongoDB Vector Search is entering public preview. The feature generates embeddings automatically when data is written or updated, helping AI systems retrieve more current information without requiring developers to build separate embedding pipelines.

This matters because AI agents depend on both memory and retrieval. Embeddings turn data into vectors so systems can find related information based on meaning rather than exact wording. The feature removes a layer of manual work that has often sat between a company's data and its AI search tools.

Another part of the update is the LangGraph.js Long-Term Memory Store, now generally available. It gives JavaScript and TypeScript developers persistent memory across conversations using MongoDB Atlas as the backend, extending a feature previously available to Python developers to a wider set of programming environments.

CJ Desai, President and Chief Executive Officer of MongoDB, said the company sees data infrastructure rather than model design as the main obstacle to deploying AI agents reliably. "The hardest part of running agents in production isn't the model. It's the data layer underneath it," Desai said.

"To trust an agent at scale, it has to retrieve the right context, hold memory across sessions, and operate at machine speed, wherever the enterprise needs it. That's why AI-native companies like ElevenLabs build voice agents on MongoDB, and why institutions like Lloyds Banking Group trust it for mission-critical workloads," he said.

Database changes

MongoDB also released version 8.3 of its database software. The update delivers up to 45% more reads, 35% more writes, 15% more ACID transactions and 30% more complex operations than MongoDB 8.0, without requiring application code changes.

Those changes are paired with new query expressions for data transformation, moving some common processing tasks into the database itself. That could reduce the need for external pipelines used to prepare data before feeding it into AI systems.

Pablo Stern, Chief Product Officer, AI and Emerging Products at MongoDB, said the company wants developers to spend less time managing infrastructure. "When AI tools and agents produce a wrong answer, the instinct is to blame the model," Stern said.

"But the data platform is what enables the agent with the right context and memory to act correctly. With MongoDB, we've made this easy. Developers no longer have to build and maintain data infrastructure, wire up embeddings, or manage syncing between systems. They can focus on business outcomes rather than the plumbing," he said.

Private networks

MongoDB also added cross-region connectivity for AWS PrivateLink, now generally available. The feature keeps traffic between Atlas clusters in different AWS regions on Amazon's private network rather than exposing it to the public internet.

That is likely to matter most for sectors such as banking, healthcare and government, where network design is often shaped by data residency rules and internal security requirements. The update is intended to make it easier for customers to approve cross-region architectures while maintaining private connectivity.

The broader pitch is that customers should be able to run the same database system across Amazon Web Services, Google Cloud, Microsoft Azure, on-premises systems and hybrid environments. MongoDB says this gives customers one database, one API and one common operating model across those deployment choices.

Ben Cefalo, Chief Product Officer, Core Products at MongoDB, linked the latest database release to rising demand from large organisations running AI services at scale. "The requirements of enterprises running AI at scale are what we build for. MongoDB 8.3 makes agent workloads faster and cheaper to run on infrastructure customers already have. We've also moved common data transformations into the database itself, so teams no longer have to maintain external pipelines just to feed their agents. Production AI doesn't wait, and neither do we," Cefalo said.

Alongside the headline changes, MongoDB also announced a Feast feature store integration and AI skill badges, both generally available. Customers including ElevenLabs and Lloyds Banking Group are already using MongoDB for AI and other critical workloads, the company said.