Amazon Bedrock AgentCore Runtime provides a cloud-based, isolated execution environment for coding agents, allowing developers to offload tasks from their local laptops. By utilizing isolated Linux microVMs, the system ensures each agent session maintains a persistent workspace, a functional shell, and deterministic command execution. This setup addresses common limitations of local development, such as agent suspension when a laptop lid is closed, security risks from sharing local credentials, and process collisions when running multiple agents simultaneously.

The platform enables developers to run coding agents like Claude Code, Codex, Kiro, Cursor, and Gemini CLI in parallel across separate environments. Each session includes managed storage through a persistent /mnt/workspace directory that survives for 14 days of inactivity, eliminating the need for manual file synchronization. Starting June 5th, AgentCore Runtime added support for interactive PTY-backed shells, allowing terminal access to these remote microVMs. Users can reconnect to sessions or shells even after network drops or system reboots, maintaining the state of their working directory and scrollback.

For platform teams, AgentCore Gateway centralizes tool and credential management to maintain security standards. Rather than placing secrets like GitHub or AWS credentials directly within the agent's file system, Gateway acts as an interface that manages tool access through an MCP (Model Context Protocol) endpoint. Credentials remain protected in AWS Secrets Manager or a centralized Token Vault, and access is managed through IAM or OAuth 2.0. This allows agents to perform tasks like opening pull requests or interacting with Jira and Slack while ensuring all actions are logged via AWS CloudTrail and Amazon CloudWatch.

Amazon Bedrock AgentCore Runtime provides a cloud-based, isolated execution environment for coding agents, allowing developers to offload tasks from their local laptops. By utilizing isolated Linux microVMs, the system ensures each agent session maintains a persistent workspace, a functional shell, and deterministic command execution. This setup addresses common limitations of local development, such as agent suspension when a laptop lid is closed, security risks from sharing local credentials, and process collisions when running multiple agents simultaneously.

The platform enables developers to run coding agents like Claude Code, Codex, Kiro, Cursor, and Gemini CLI in parallel across separate environments. Each session includes managed storage through a persistent /mnt/workspace directory that survives for 14 days of inactivity, eliminating the need for manual file synchronization. Starting June 5th, AgentCore Runtime added support for interactive PTY-backed shells, allowing terminal access to these remote microVMs. Users can reconnect to sessions or shells even after network drops or system reboots, maintaining the state of their working directory and scrollback.

For platform teams, AgentCore Gateway centralizes tool and credential management to maintain security standards. Rather than placing secrets like GitHub or AWS credentials directly within the agent's file system, Gateway acts as an interface that manages tool access through an MCP (Model Context Protocol) endpoint. Credentials remain protected in AWS Secrets Manager or a centralized Token Vault, and access is managed through IAM or OAuth 2.0. This allows agents to perform tasks like opening pull requests or interacting with Jira and Slack while ensuring all actions are logged via AWS CloudTrail and Amazon CloudWatch.