backyard 0.2.0

Fast Python sandboxes for AI agents.

Backyard

Fast Python Sandbox for AI agent harnesses.

backyard supports transparent execution of Python code using either:

• Pydantic-Monty: an ultra-fast, minimal Python interpreter
• a container engine (Docker or Podman): full-featured Python support

Getting started

Installation

Using uv (recommended):

uv add backyard

Or, with pip:

pip install backyard

Python API

from pprint import pprint

from backyard import Sandbox

sandbox = Sandbox()

code = """
print("hello from the sandbox! 🏖🏰")
"""

result = sandbox.run(code)

pprint(result)

Security

gVisor (runsc) sandboxing

For maximum isolation when running untrusted code, backyard automatically detects and uses gVisor (runsc) as the container runtime if it is registered with your container engine. gVisor provides an additional kernel-level security boundary between the sandboxed code and the host system.

If gVisor is not detected, backyard falls back to the default runc runtime and logs a warning.

Installing gVisor

1. Install the runsc binary:

   # Ubuntu / Debian
   sudo apt-get install runsc
   
   # Or download the latest release
   wget https://storage.googleapis.com/gvisor/releases/release/latest/runsc
   sudo mv runsc /usr/local/bin/runsc
   sudo chmod +x /usr/local/bin/runsc
   

2. Register runsc with your container engine by editing the daemon configuration:

   Docker — add to /etc/docker/daemon.json:

   {
       "runtimes": {
           "runsc": {
               "path": "/usr/local/bin/runsc"
           }
       }
   }
   

   Podman — add to ~/.config/containers/containers.conf or /etc/containers/containers.conf:

   [engine.runtimes]
   runsc = ["/usr/local/bin/runsc"]
   

3. Restart the container engine:

   # Docker
   sudo systemctl restart docker
   
   # Podman (no daemon, but verify the config is picked up)
   podman info --format '{{json .Host.OCIRuntimes}}'
   

Once configured, backyard will automatically detect and use gVisor for all container sandboxes, providing an additional kernel-level security boundary.

FAQ

Why did you create this?

Originally, I built the dual-engine sandbox for fine-tuning large language models using Reinforcement Learning from Verifiable Rewards (RLVR) with Group Relative Policy Optimization (GRPO). The sandbox provided an environment to safely execute model-generated code and generate rewards for training the model.

I then expanded the sandbox capabilities for use in the my-ai coding agent harness I am building.

Where does the name backyard come from?

The term "sandbox" refers to a controlled environment where untrusted code can be run safely.

In the physical world, a sandbox is a place where children explore and imagine as they create and shape the world around them. 🏖🏰

Where does one place a sandbox? In the backyard.