Skip to content

A Python interpreter with built-in safeguards for executing untrusted code, like LLM-generated scripts.

License

Notifications You must be signed in to change notification settings

aremeis/pytherpreter

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🐍 pytherpreter 🐍

A Python interpreter with built-in safeguards for executing untrusted code, like LLM-generated scripts.

This repository contains the Python interpreter tool extracted from HuggingFace’s smolagents project. Big hug to the HuggingFace team for their initial implementation! 🤗

Some improvements over the smolagents tool:

  • Supports async code execution using the async_evaluate function and AsyncPythonInterpreter class.
  • Improved function call resolution.
  • Supports custom subscriptable objects.
  • No external dependencies.
  • More flexible print handling.

Installation

pip install pytherpreter

Latest development version:

pip install git+ssh://git@github.com/aremeis/pytherpreter.git

or

pip install git+https://github.com/aremeis/pytherpreter.git

Usage

Using evaluate

This function evaluates Python code and returns the result.

from pytherpreter import evaluate

result = evaluate("""
from math import sqrt
sqrt(4)
""")
print(result)

# Output:
# 2.0

The evaluate function returns the result of the last expression in the code.

Using PythonInterpreter

This class is a wrapper around the evaluate function that keeps the state of the interpreter between calls. Variables and functions defined by the code will be be available in subsequent calls.

from pytherpreter import PythonInterpreter

interpreter = PythonInterpreter()
result = interpreter("x = 3")
print(result)

# Output:
# 3

result = interpreter("x += 1")
print(result)

# Output:
# 4

Printing

You may provide a stdout argument to capture the output of print statements in the code.

from pytherpreter import evaluate
import io
stdout = io.StringIO()
result = evaluate("print('Hello, World!')", stdout=stdout)
print(stdout.getvalue())

# Output:
# Hello, World!

Variables

You may provide a variables argument to preset variables and capture changes to them.

from pytherpreter import evaluate
variables = {"x": 3}
result = evaluate("x += 1", variables=variables)
print(variables["x"])

# Output:
# 4

Safeguards

Built-in functions

You may provide a builtin_functions argument containing a dictionary of built-in functions the code is allowed to call. If you don't provide this argument, the code will only be able to call the built-in functions in BASE_BUILTIN_FUNCTIONS.

The code will not be able to modify the provided built-in functions.

Modules

By default, the code will only be able to import the modules in BASE_BUILTIN_MODULES. You may provide an authorized_imports argument to allow the code to import additional modules.

Documentation

For more details, the reference documentation is available here.

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.

About

A Python interpreter with built-in safeguards for executing untrusted code, like LLM-generated scripts.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages