LineRun

LineRun is an isolated Python execution environment ("sandbox") tailored specifically for autonomous AI agents. It provides a clean, path-traversal-protected interface for writing code, managing dependencies via virtual environments, and executing Python scripts safely.

Features

• Virtual Environment Isolation: Automatically creates a dedicated venv for each runner instance.
• Path Traversal Protection: Enforces strict boundaries to ensure agents cannot read, edit, or execute files outside the designated sandbox directory.
• AI-Friendly File I/O: Provides methods specifically designed to reduce LLM token usage and formatting hallucinations, including full-file overwrites and exact-string replacements.
• Dependency Management: Native API for pip installing and managing external modules within the sandbox.
• NixOS Compatibility: Automatically detects NixOS and wraps execution in a nix-shell configured to properly link dynamic C/C++ libraries (like libstdc++, OpenGL, X11) when installing binary wheels from PyPI.

Installation

LineRun is built using standard Python libraries and can be packaged via PyPI.

git clone 
cd LineRun
pip install -e .

If you intend to run this in a production environment with untrusted AI code, it is highly recommended to install and run LineRun inside an isolated Docker container. Directory separation alone does not prevent malicious OS-level system calls or network access.

Quick Start

from linerun.main import Code_Runner

# 1. Initialize the sandbox
# This automatically creates the venv
runner = Code_Runner(Path="./my_sandbox", Venv_Path="./my_venv")

# 2. Add dependencies
runner.Add_Module("requests")

# 3. Write code
script = '''
import requests
resp = requests.get("https://api.github.com")
print("Status Code:", resp.status_code)
'''
runner.Write_File("fetch.py", script)

# 4. Execute the code
output = runner.Run_Code("fetch.py")
print("Agent Output:", output)

API Reference

Initialization

• Code_Runner(Path, Venv_Path): Creates a new sandboxed runner. Will throw an error if Path is not empty. Automatically scaffolds the Python virtual environment.

File Operations

• Write_File(Name, Content): Overwrites (or creates) a file with the exact content provided.
• Replace_In_File(Name, TargetContent, ReplacementContent): Finds the first exact match of TargetContent and replaces it. Ideal for surgical patches. Throws an error if the match is not exact.
• Read_File(Name): Returns the string contents of a file.
• Add_File(Name): Creates an empty file.
• Delete_File(Name): Deletes a file.
• All_files(regex=None): Returns a list of all files relative to the sandbox path. Supports optional regex filtering.

Dependency Management

• Add_Module(Name): Uses pip to install a package into the virtual environment.
• Remove_Module(Name): Uninstalls a package.
• List_Modules(): Lists all installed pip packages.

Execution

• Run_Code(Name): Executes a file using the virtual environment's Python binary. Captures and returns standard output.