AWS says Amazon Bedrock AgentCore Runtime can host coding agents such as Claude Code, Codex, Kiro, OpenCode, Cursor CLI, and Gemini CLI in isolated Firecracker microVM sessions, with persistent filesystems, secure tool access, and support for long-running work.
AWS has published guidance on hosting coding agents with Amazon Bedrock AgentCore Runtime, positioning the service as a way to keep agentic development tasks running in the cloud even after a developer closes a laptop.
In an AWS Machine Learning Blog post titled “It’s safe to close your laptop now: Hosting coding agents on Amazon Bedrock AgentCore,” AWS describes running coding agents such as Claude Code, Codex, Kiro, OpenCode, Cursor CLI, and Gemini CLI on Amazon Bedrock AgentCore Runtime.
The post focuses on a common limitation of local coding-agent workflows: long-running tasks can be interrupted when a developer’s machine sleeps, disconnects, or loses access to required tools. AWS says AgentCore Runtime addresses this by hosting coding-agent sessions in the cloud while maintaining access to a persistent workspace and approved tools.
According to the AWS blog, the setup is intended for coding assistants that can inspect repositories, modify files, run commands, and interact with external services through controlled interfaces. AWS describes these agents as running inside isolated Firecracker microVM sessions, rather than directly on a developer’s laptop.
AWS Documentation for “Host agent or tools with Amazon Bedrock AgentCore Runtime” says the service provides a secure serverless hosting environment for agents and tools. The documentation states that each user session runs in a dedicated microVM, which is designed to isolate sessions from one another.
The same AWS documentation says AgentCore Runtime supports persistent filesystems, allowing state and files to remain available across a session. It also says the runtime supports long-running workloads of up to eight hours. Those capabilities are central to the coding-agent use case described in the AWS blog, where an agent may need time to analyze code, make changes, run tests, or coordinate with developer tooling.
AWS also emphasizes secure tool access. In the coding-agent scenario, that can include access to repositories, command-line tools, and external services through configured connections rather than unrestricted local-machine access.
AWS also points to a companion sample in the awslabs GitHub repository under “Coding Agents on Amazon Bedrock AgentCore.” The sample describes deploying coding agents including Claude Code, Kiro, Codex, Cursor, Hermes, and OpenCode on AWS Bedrock AgentCore.
The GitHub sample includes a GitHub MCP Gateway and a comparison frontend, according to its description. MCP, or Model Context Protocol, is used in many agent workflows to connect models and agents with external tools and data sources through a defined interface. In this sample, AWS presents the gateway as part of an end-to-end environment for comparing multiple coding agents hosted on Bedrock AgentCore.
The sample does not mean every listed coding tool has the same capabilities or deployment model. Rather, the repository shows how different coding-agent CLIs and related tools can be hosted in the AgentCore Runtime pattern described by AWS.
For developers experimenting with coding agents, the AWS materials highlight three practical concerns: keeping work running beyond the local laptop session, isolating agent execution, and preserving project state during longer tasks.
A cloud-hosted coding agent can continue operating while the developer is away, but AWS’s documentation also makes clear that the runtime is still bounded: it supports long-running workloads up to eight hours, not indefinite execution. Teams adopting this model would still need to manage credentials, repository permissions, tool approvals, auditability, and cost.
The AWS blog and documentation present AgentCore Runtime as infrastructure for hosting agents and tools rather than as a single coding agent. That distinction matters: Bedrock AgentCore provides the runtime environment, while tools such as Claude Code, Codex, Kiro, OpenCode, Cursor CLI, and Gemini CLI remain separate agent or command-line systems that can be deployed within that environment.
The result is a model in which coding agents can run in isolated cloud sessions with persistent workspace support and controlled access to development tools. AWS is framing that as a way to make agentic coding workflows more resilient than laptop-bound sessions, while keeping execution within a managed AWS runtime.
AWS has published guidance on hosting coding agents with Amazon Bedrock AgentCore Runtime, positioning the service as a way to keep agentic development tasks running in the cloud even after a developer closes a laptop.
The post focuses on a common limitation of local coding agent workflows: long running tasks can be interrupted when a developer’s machine sleeps, disconnects, or loses access to required tools.
AWS says AgentCore Runtime addresses this by hosting coding agent sessions in the cloud while maintaining access to a persistent workspace and approved tools.
Continue reading
