Sandbox Runtime
A lightweight operating-system-level sandbox tool open-sourced by Anthropic. It intercepts and controls AI agents' access to the file system and network at the system-call level, requires no containers or virtualization, and supports macOS and Linux.
Timeline (last 90 days)
The effectiveness of the sandbox depends on how strict the configuration is; users must explicitly allow resources and paths, otherwise they may get a false sense of security.
Sandbox Runtime requires no Docker or containers and runs directly on the system.
Sandbox Runtime currently supports Mac and Linux systems.
Sandbox Runtime does not yet support Windows, and Windows users will have to wait.
Sandbox Runtime is a free sandboxing tool released by Anthropic, and it powers the underlying sandboxing technology of Claude Code.
Sandbox Runtime is open source under the Apache 2.0 license.
Sandbox Runtime has gained more than 6,000 stars on GitHub.
The core function of Sandbox Runtime is to limit the scope of what AI agents can do: agents cannot access websites that are not explicitly allowed, nor can they read private files (such as SSH keys).
Sandbox Runtime is installed with a single npm command.
When using Sandbox Runtime, the sandbox is enabled by prefixing any command with `srt`, for example `srt claude`.
3 more timeline events
All Facts (13)
The effectiveness of the sandbox depends on how strict the configuration is; users must explicitly allow resources and paths, otherwise they may get a false sense of security.
50%UnverifiedSandbox Runtime requires no Docker or containers and runs directly on the system.
50%UnverifiedSandbox Runtime currently supports Mac and Linux systems.
50%UnverifiedSandbox Runtime does not yet support Windows, and Windows users will have to wait.
50%UnverifiedSandbox Runtime is a free sandboxing tool released by Anthropic, and it powers the underlying sandboxing technology of Claude Code.
50%UnverifiedSandbox Runtime is open source under the Apache 2.0 license.
50%UnverifiedSandbox Runtime has gained more than 6,000 stars on GitHub.
50%UnverifiedThe core function of Sandbox Runtime is to limit the scope of what AI agents can do: agents cannot access websites that are not explicitly allowed, nor can they read private files (such as SSH keys).
50%UnverifiedSandbox Runtime is installed with a single npm command.
50%UnverifiedWhen using Sandbox Runtime, the sandbox is enabled by prefixing any command with `srt`, for example `srt claude`.
50%UnverifiedThe four tools form a layered protection system: Skill Spectrum and Cisco MCP Scanner handle pre-installation static detection, Agent Scan handles inspection of already-installed content, and Sandbox Runtime provides an isolation barrier at runtime.
50%UnverifiedAnthropic has open-sourced sandbox-runtime (srt), the OS-level sandbox behind Claude Code, used to constrain an agent's file and network access without requiring a container.
50%UnverifiedAnthropic's Sandbox Runtime provides runtime-level isolation protection, confining agents to an independent folder.
50%