Amazon Bedrock Unveils Stateful Runtime Environment for Agents

Existing Stateless API-based agent prototypes work well for single prompts or simple tool calls, but they fall short in real production environments. Real-world workflows span multiple steps and require maintaining context from previous tasks, tool outputs, approval processes, and system state.

Development teams have had to bear the burden of building separate orchestration layers themselves to handle state persistence, tool invocation, error handling, and resumption of long-running tasks.

The Stateful Runtime Environment, born from the collaboration between OpenAI and Amazon, is designed to run natively within Amazon Bedrock to reduce this burden. Built on OpenAI models, it provides the following core capabilities:

• Working Context persistence: automatically manages memory/history, tool and workflow state, environment usage information, and permission boundaries
• Automatic execution of complex steps: agents perform complex steps on their own without needing to manually chain together scattered requests
• AWS-native integration: operates in compliance with existing AWS security policies, tools, and governance rules

This enables enterprises to deploy complex workflows requiring approval and auditing—such as customer support, sales operations, IT automation, and financial processes—into production more quickly.