MongoDB has unveiled a major expansion of its data platform, introducing a system for deploying and managing AI agents along with a new database architecture built to handle sudden shifts in compute demand.

The releases include Atlas Agent Engine, now in public preview, MongoDB 9.0, which is generally available, and Atlas Infinite, also in public preview. In combination, the products extend MongoDB well beyond its roots as a document database provider, with a focus on running AI applications using real-time business data.

The most AI-focused offering is Atlas Agent Engine, which combines agent execution with persistent memory, data retrieval and governance. The goal is to give development teams a common environment for moving agents from experimental deployments into production without needing numerous third-party tools.

Atlas Agent Engine logs actions to an identity, whether an agent or a human, and applies governance policies through a common control plane. Its memory capabilities can be deployed separately from the runtime, allowing companies to use existing AI models and development frameworks.

Open Approach to AI Models

MongoDB does not require customers to commit to a single model or cloud. The platform supports standards including MCP and A2A and is designed to work with different AI models and frameworks. Retrieval uses technology from Voyage AI, which MongoDB acquired in 2025.

MongoDB 9.0 addresses the database demands created by AI workloads. According to the company, it delivers up to 35% faster find-one queries and up to 30% faster update-one queries compared with version 8.0. MongoDB also claims up to twice the throughput on large instances and up to 20% greater throughput for transactional workloads.

Version 9.0 adds the ability to perform prefix, suffix and substring searches on encrypted information without first decrypting the data. Other updates include workload management intended to protect clusters during heavy demand, enhanced backup and recovery features and expanded OpenTelemetry support.

As for Atlas Infinite, its architecture separates compute and storage so the two resources can expand independently. This approach supports the actions of AI agents, whose activity can be difficult to predict. Large numbers of agents can generate a rapid burst of database reads and writes rather than the more predictable traffic of traditional applications.

MongoDB said its internal testing found Atlas Infinite delivered 189% more throughput per dollar than Atlas Core when running MongoDB 9.0.

These new product releases place MongoDB in a crowded enterprise AI market. Database companies including Redis and Couchbase have expanded their AI capabilities, while data platform vendors such as Snowflake and Databricks have invested heavily in AI development.