AWS adds Bedrock web search & DynamoDB vector search

Amazon Web Services has introduced a set of product updates focused on artificial intelligence tools, data services and serverless computing. The changes include new web search access in Amazon Bedrock.

The latest announcements also cover dedicated runtime instances for Bedrock AgentCore, vector search in DynamoDB, broader availability for AWS Transform continuous modernisation and higher network bandwidth for some AWS Lambda functions. Together, they show AWS focusing on developer tools for AI agents and updates to existing cloud services rather than launching a single standalone platform.

Bedrock changes

The most prominent update is web search in Amazon Bedrock. OpenAI models available through Bedrock, including GPT-5.4, GPT-5.5 and GPT-5.6 Sol, Terra and Luna, can now browse and retrieve information from the internet, giving applications access to current information beyond model training data.

The feature is intended for AI applications and agents that need answers based on live web content while remaining inside a customer's AWS environment. AWS said the setup is designed to maintain data residency with no data egress.

Runtime instances on Amazon Bedrock AgentCore are also now generally available. The service lets customers deploy and run AI agents on dedicated runtime instances rather than shared infrastructure, offering more control over execution environments and more predictable costs, according to AWS.

The focus on Bedrock reflects a broader shift among large cloud providers to make managed AI services more practical for production use. Rather than competing only on access to frontier models, vendors are increasingly adding retrieval, orchestration, governance and deployment controls around those models.

Database and code

DynamoDB now supports vector search, allowing users to store and query vector embeddings alongside existing application data in the database. In some cases, that removes the need to maintain a separate vector database for semantic search or retrieval tasks.

The update is particularly relevant for AI agent developers using DynamoDB to store memory or application state. By adding vector search to the same service, AWS is aiming to reduce the operational complexity of stitching together multiple databases for transactional data and semantic retrieval.

AWS Transform continuous modernisation is also now generally available. The service is intended to help engineering teams analyse and remediate technical debt across source code repositories at scale, including work linked to legacy and mainframe systems.

The launch suggests AWS sees modernisation as an ongoing software maintenance task rather than a single migration project. That framing aligns with broader demand from large organisations trying to reduce the risks and costs tied to ageing applications without undertaking one-off rewrites.

Lambda bandwidth

For serverless users, Lambda now supports network bandwidth of up to 3,000 Mbps for functions outside a virtual private cloud when configured with 2 GB of memory or more. Bandwidth scales from 625 Mbps at 2 GB to 3,000 Mbps at 10 GB, according to AWS.

That change is likely to matter for workloads that move large volumes of data between Lambda and other AWS services. It may also make Lambda more suitable for some data-heavy processing jobs that have previously pushed customers towards container- or instance-based alternatives.

Open source and standards

AWS has also open-sourced Dogwood, which it described as a governance language for AI agents. The project is intended to support Cedar policies and add temporal conditions, extending policy decisions beyond a single request to include the history of an agent's actions within a session.

The company also announced support for Agent Plugins, which it described as an open-source, vendor-neutral specification for packaging agent extensions. AWS said the format is designed so an extension can be packaged once and then used across clients including Kiro, VS Code, Cursor and other tools that implement the specification.

Those moves suggest AWS is trying to position itself not only as a provider of managed AI infrastructure but also as a participant in emerging standards for how software agents are governed and extended. Open standards remain a key issue for customers wary of becoming locked into one vendor's tooling as agent-based software development evolves.

AWS also highlighted Kiro Crew, a workspace for collaborative multi-agent development workflows inside the Kiro integrated development environment. The product is described as persistent and able to support several streams of work across repositories, tools and longer periods rather than a single chat interaction.

The range of updates underlines how much of AWS's current AI strategy is built around agent development, model grounding, governance and developer workflow. It also shows the company extending familiar services such as DynamoDB and Lambda to keep them relevant as customers look for ways to build AI applications without overhauling the rest of their cloud stack.