shuru.run

What it is

shuru is a local-first microVM sandbox tool built for AI Agents. It is positioned as a lightweight Linux VM execution environment for macOS Apple Silicon and Linux ARM64. Its key idea is “Ephemeral by default”: every run starts from a clean rootfs, and changes are not preserved after exit by default, making it suitable for running untrusted or AI-generated code.

Core capabilities

Functionally, shuru focuses on security, reproducibility, and automation. Its CLI can start a sandbox, run commands, and destroy the environment with a single command. Networking is disabled by default, can be enabled via --allow-net, and domain allowlists can be configured in a config file. Resource limits can be set for CPU, memory, and disk size. Directory mounts are read-only by default; writes back to the host are only allowed with explicit :rw plus --allow-host-writes. Its checkpoint mechanism can save disk state, restore it, and branch from it, similar to making a git commit for an environment. It also supports port forwarding, allowing guest services to be exposed on the local machine even when general networking is not enabled.

Integrations, platforms, and documentation

In terms of platform support, the main documentation explicitly supports macOS Apple Silicon via Apple Virtualization.framework, while Linux ARM64 uses KVM. There is no mention of Windows, x86 Linux, or cloud-hosted options. In the ecosystem, it provides agent skills and can work with Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenAI Codex, and others. Installation methods include Homebrew, a script, and npx skills add. The documentation includes many CLI and shuru.json configuration examples, making onboarding clear, but it lacks sufficient information on licensing, open-source status, commercial support, and APIs/SDKs.

Pricing, strengths, and limitations

The page does not disclose pricing, payment methods, or commercial plans. Its strengths include offline-by-default behavior, no Docker dependency, fine-grained permission control, and a practical checkpoint feature, making it especially suitable for Agents that need to execute code and install dependencies. Its limitations are narrow platform coverage and missing information on enterprise management, auditing, remote collaboration, SDKs, and related areas.

Who it is for and access from China

It is suitable for AI Agent developers, code execution platforms, evaluation systems, and engineers who need disposable Linux environments. Access from China cannot be determined from the main text. If it depends on GitHub, Homebrew, npm, or external model APIs, actual installation and usage may be affected by network conditions. Alternatives to consider include Docker, Firecracker, Lima, Colima, Devbox, and GitHub Codespaces.