Volume 343, Issue 3 · 7 articles from 2 issues
MongoDB shows how to encrypt LangGraph agent checkpoints with Client-Side Field Level Encryption.
Plaintext agent checkpoints can be read by database administrators or anyone who gains database access in a breach, and this MongoDB post shows how to use Client-Side Field Level Encryption to encrypt LangGraph checkpoints before they are written to the database.
The application decrypts state when resuming a run, while anyone without the keys sees only ciphertext.
The example extends LangGraph's CipherProtocol and stores the encrypted payload through MongoDBSaver.
MongoDB explains how retailers can enrich product catalogs so AI shopping agents can find and compare products.
MongoDB says retail discovery is shifting from clicks and filters to conversations that AI agents conduct for customers.
AI recommendations will favor products that can be systematically retrieved, understood through rich context and compared consistently, rather than those with the most persuasive descriptions.
The post cites McKinsey's estimate that AI agents could orchestrate $3 trillion to $5 trillion in commerce by 2030, and shows how retailers can enrich catalogs with MongoDB.
MongoDB's open-source Hibernate ORM extension is GA, adding JOINs, optimistic locking and Spring Boot support.
The MongoDB Extension for Hibernate ORM is now generally available as a first-party, open-source library that lets Java applications use MongoDB through Hibernate, so developers can keep using @Entity annotations, HQL, Criteria Queries and Session APIs for new apps or when migrating relational Hibernate workloads.
The GA release adds INNER and LEFT OUTER JOINs, optimistic locking, Spring Boot auto-configuration and clearer exceptions and logging.
It requires Hibernate ORM 7.4 LTS.
MongoDB shows how to build a knowledge assistant with Grok Build's coding agents and MongoDB Atlas.
MongoDB says agentic platforms like Grok Build have redefined the unit of software work, with developers giving autonomous agents goals, constraints, a codebase and a definition of done.
The tutorial shows how to combine that workflow with MongoDB Atlas as the operational database, memory layer and retrieval engine for a small knowledge assistant.
It recommends keeping rules, specs, tests, seed data and deployment files in Git so humans and agents share one source of truth.
The post also explains how coding agents use MCP to work with external tools.
MongoDB launches MongoDB Agents, open source agent templates for routine database work on Atlas Agent Engine.
MongoDB says building agents for routine database work requires credential handling, access scoping, tool orchestration, approval flows, chat integration, evaluation and observability.
MongoDB Agents is a collection of open source templates that handle this plumbing by combining the MongoDB MCP Server and MongoDB Agent Skills with scoped Atlas credentials, write approval flows and ready-to-run deployment configuration.
Users bring their own model API key and deploy to Atlas Agent Engine.
It launches with two templates: Atlas Explorer for conversational workflows and Atlas Query Insights for slow-query reports.
MongoDB releases AptMQL-Bench, a text-to-MQL benchmark with 21 databases and 3,186 natural-language requests.
MongoDB says LLMs increasingly translate requests into database queries, but its query language MQL has received far less benchmark attention than SQL.
It introduces AptMQL-Bench, a text-to-MQL benchmark built by converting existing text-to-SQL datasets from BIRD.
The release includes 21 new MongoDB databases with 3,186 natural-language requests across domains such as commerce, science, sports and finance.
MongoDB also publishes a technical report on its agent-driven conversion pipeline, the prompts and an evaluation script with a soft execution accuracy metric.
MongoDB begins a series on building an agentic warehouse management system, identifying where AI agents add value.
A MongoDB author draws on years in banking and logistics to explore how an AI agent can become part of a real warehouse management workflow.
The project is not meant to be a complete WMS but a familiar logistics scenario for testing where agents help.
Part 1 introduces the scenario and identifies opportunities for an agent.
Later parts will connect the agent with Java and Spring AI and then execute replenishment decisions.