AWS unveils OpenSearch Serverless, a database built for AI agents

AWS has announced a new version of OpenSearch Serverless, a managed database designed from the ground up for AI agents. It is a landmark moment: one of the world's largest cloud providers is effectively acknowledging that the architecture of the internet is changing radically.

What OpenSearch Serverless can do

The service is built specifically for agentic workloads, where autonomous AI systems must quickly search, process and analyze large volumes of text and vector data. OpenSearch Serverless supports vector search for semantic analysis and full-text keyword search, and it runs in no-code mode: developers do not need to manage a server or write configuration files to initialize it.

The service scales automatically with load and requires exactly zero upfront setup. That is critical for agents that must run 24/7 unattended and adapt to unpredictable traffic spikes.

What problems it solves

OpenSearch Serverless is versatile. Agents use it for:

• Finding relevant information in their knowledge base before making a decision
• Analyzing structured data and documents in real time
• Processing large volumes of text and synthesizing information from many sources
• Working with vector representations of data for semantic search, where meaning matters more than keywords

This matters because modern AI agents differ sharply in architecture from a typical web application. They need instant access to knowledge, can process hundreds of requests in parallel and, critically, run at the customer's expense: they are billed not by working time but by the number of operations and the volume of data.

The internet is being rebuilt for machines

The event is symbolic. Just five years ago, cloud infrastructure was built around peak loads from mobile apps and websites. Today AWS, Google Cloud and Azure are all releasing services fully optimized for AI. Amazon has reworked its Trainium chips for training large language models. Meta and Google are building their own custom processors for inference. OpenAI and Anthropic require gigawatts of electricity to run their models. In parallel, cloud providers are redesigning databases, caching systems and monitoring systems, everything agents might need to operate autonomously.

Users and their data are moving to the background. The internet's core architecture is now being rebuilt around the needs of machines that must talk to each other quickly and without delay.

What this means

The traditional internet was built for people who visit websites, watch videos and write emails. The machine internet is built for agents that dance with each other at a pace beyond human perception. OpenSearch Serverless is one more brick in the architecture of that new network, where people increasingly become observers and code does the main work.

*Meta is designated an extremist organization and banned in Russia.