Overview
Amazon Bedrock is the agent-orchestration service in the AWS architecture described by the source, handling user requests and routing them to Lambda action groups.1 The example pairs a Bedrock agent with a DynamoDB table that serves as both the operational store and vector store, allowing the same system to answer conversational questions and perform direct document operations.1
Agent Actions
The architecture assigns separate actions to Lambda functions: one performs semantic retrieval and another carries out create, read, update, and delete operations against DynamoDB.1 For semantic retrieval, the Lambda function obtains a query embedding using Amazon Titan Text Embeddings V2 and calls DynamoDB’s SearchVectors API.1 This divides responsibility between the agent’s request routing and the functions that perform database work.1
Data Synchronization
The design uses DynamoDB Streams to trigger an embedding pipeline when stored content changes, helping keep vectors aligned with their source records without manual updates.1 Keeping document data and embeddings in the same database is presented as a way to reduce the synchronization burden and the chance of inconsistent retrieval results.1
Use Case
The pattern is presented for knowledge-management systems where users need natural-language conversational search as well as direct access to documents.1 Its broader architectural motivation is to simplify an agent stack by reducing the need to operate and synchronize a separate vector database.1
Security considerations
A reported Bedrock AgentCore issue showed how prompt-driven access to the Instance Metadata Service could expose temporary cloud credentials: the service initially used IMDSv1, and weak network isolation in its Firecracker microVM setup enabled an SSRF-style request to the credential endpoint.2 The credentials reportedly enabled cross-agent discovery, access to ECR images, agent memories and conversations, and retrieval of Secrets Manager secrets; Zenity also identified permissions scoped broadly across regional AgentCore resources.2 AWS later moved AgentCore to IMDSv2 and addressed the permission issue, while disputing the researchers’ characterization of the incident as a vulnerability.2
- Dec 2025
- Jan 2026
- 22 Jun 2026
- 29 Sept 2026
“Researchers at Zenity Labs disclosed their findings to AWS in December 2025.”
“The Zenity researchers told Bob's tale, or something like it, to AWS last December, then followed up in January 2026 with details about AgentCore being overprivileged.”
“AgentCore's excessive permissions remained in place at least until June 22, 2026, when Zenity checked in and found the issue hadn't been remediated.”
“A final review by Zenity occurred on September 29, 2026, at which point the biz observed that AWS had addressed the extant problems, clearing the way for Bob's hypothetical adventure to finally be told.”
Usage
In an AWS reference architecture for antimicrobial-resistance surveillance, Bedrock Guardrails are used in the conversational query layer, while an AgentCore-hosted Strands agent calls Athena-backed tools to answer users' natural-language questions.23
FleetGPT example
In Novus Hi-Tech’s FleetGPT, Amazon Bedrock Knowledge Bases indexes unstructured standard operating procedures, policies, and documentation so the system can retrieve them during natural-language fleet-data investigations.24
Related Pages
- Amazon Bedrock AgentCore — AgentCore extends AWS agent capabilities with runtime isolation and permission safeguards alongside Amazon Bedrock’s managed foundation-model and agent services.
- FleetGPT — FleetGPT uses Amazon Bedrock’s managed foundation models and agent capabilities to investigate fleet events and retrieve operational policy documents.
- Amazon DynamoDB — DynamoDB complements Amazon Bedrock in FleetGPT by providing operational records and vector search for fleet-event access and retrieval.