Repolode
Submit
vndee

vndee/llm-sandbox

Python library that runs LLM-generated code in isolated sandboxes through one API across Docker, Kubernetes, Podman and local Linux or Windows backends.

View on GitHubWebsite
Stars
1,144
Forks
106
Open issues
25
Last commit
yesterday

What is it?

LLM Sandbox provides a common Python API for executing code produced by language models inside an isolated environment. A developer picks a backend such as Docker, Kubernetes, Podman, Bubblewrap, NsJail, or experimental Windows adapters, configures resource and network controls, and then calls the same run, install and file-copy methods regardless of which backend is in use.

The library supports Python, JavaScript, Java, C++, Go and R, and can install declared libraries before execution. Artifact sessions capture generated plots, and an interactive session keeps a Python IPython kernel alive across calls for notebook-style work. A container pool option pre-warms and reuses containers for repeated executions, and an MCP server lets clients such as Claude Desktop call the sandbox directly.

The project includes examples for eleven agent frameworks, including LangChain, LlamaIndex and CrewAI. Its security policy check is advisory only: it returns a verdict but does not block execution, so callers must check code themselves before running it.

Key features

  • 01Common API across Docker, Kubernetes, Podman, Bubblewrap, NsJail and Windows adapters
  • 02Configurable CPU, memory, network and execution time limits
  • 03Supports Python, JavaScript, Java, C++, Go and R
  • 04Installs declared libraries before running code
  • 05Captures plots generated by Python and R code
  • 06Container pooling to reuse pre-warmed containers
  • 07MCP server for clients such as Claude Desktop

Use cases

  • Running code written by an LLM agent before returning results to the user
  • Adding a code interpreter tool to LangChain, CrewAI or similar agent frameworks
  • Executing untrusted scripts on a local Linux host without Docker
  • Letting a chat assistant generate and display plots safely

Quick start

  1. 1

    Install the package from PyPI

    pip install llm-sandbox
  2. 2

    Install the Docker extra for the most common backend

    pip install 'llm-sandbox[docker]'
  3. 3

    Run the basic Python example from the README with a SandboxSession

Things to know

  • Local Linux backends share the host kernel and are not a VM boundary.
  • Windows adapters are experimental, with limited language support and API features.
  • Container backends such as Docker, Kubernetes and Podman need a running container runtime.

In collections

Sandboxes for AI agentsOpen-source AI agent frameworks & apps

Topics

#llm#code-interpreter#sandbox#code-execution#docker#kubernetes#agents#python#code-generation#large-language-models

FAQ

What is llm-sandbox?
Python library that runs LLM-generated code in isolated sandboxes through one API across Docker, Kubernetes, Podman and local Linux or Windows backends. LLM Sandbox provides a common Python API for executing code produced by language models inside an isolated environment. A developer picks a backend such as Docker, Kubernetes, Podman, Bubblewrap, NsJail, or experimental Windows adapters, configures resource and network controls, and then calls the same run, install and file-copy methods regardless of which backend is in use.
Is llm-sandbox free to use? What license does it have?
llm-sandbox is open source under the MIT license. Check the license terms before using it in commercial products.
Is llm-sandbox still maintained?
Actively maintained. The last commit to llm-sandbox was on Nov 6, 2026.
What can I use llm-sandbox for?
Running code written by an LLM agent before returning results to the user Adding a code interpreter tool to LangChain, CrewAI or similar agent frameworks Executing untrusted scripts on a local Linux host without Docker Letting a chat assistant generate and display plots safely
How do I install llm-sandbox?
Install the package from PyPI: pip install llm-sandbox Install the Docker extra for the most common backend: pip install 'llm-sandbox[docker]' Run the basic Python example from the README with a SandboxSession
Where can I get llm-sandbox?
The source code and installation instructions are on GitHub: https://github.com/vndee/llm-sandbox

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects