ms-enclave Documentation

ms-enclave is a modular, robust sandbox runtime that gives your applications a secure, isolated execution environment. Backed by Docker for strong isolation, with local / HTTP managers and an extensible tool system to safely run code (including LLM-generated code) in a controlled environment.

Key features

• 🔒 Secure isolation: Docker-based isolation with resource limits
• 🧩 Modular: sandboxes, tools, and managers are all pluggable via registry/factory
• ⚡ Stable: lightweight implementation, fast startup, pool warm-up supported
• 🌐 Remote-ready: built-in FastAPI server, identical API for local and remote
• 🔧 OpenAI-compatible: tool schemas plug directly into Function Calling

Requirements

• Python ≥ 3.10
• OS: Linux / macOS / Windows (with Docker)
• A working Docker daemon; port 8888 must be free for the Notebook sandbox

Where to start

New here?

1. Installation
2. 5-Minute Quickstart — one minimal runnable example
3. Core Concepts — understand Sandbox / Manager / Tool

By task

• Pick the right entry point: Factory / Manager / Pool / HTTP
• Built-in tools
• Install third-party dependencies / Mount host directories
• Notebook sandbox (stateful)
• LLM Agent integration

Deployment

• HTTP server / HTTP client
• Volcengine cloud sandbox

Extending

• Registration overview
• Custom Tool / Sandbox / SandboxManager

API reference

• API reference (auto-generated via mkdocstrings)

Stuck?

• FAQ & troubleshooting
• File an issue: https://github.com/modelscope/ms-enclave/issues