Python Agent

[gunar]

I'd be cool if sgpt could access a python REPL (à la Langchain) to run some code in order to provide answers.

Example queries:

• What is 2023-08-01 01:00 + 54 hours?
• How much is 93 F in C?

Today, you can already pipe sgpt --code into python like this:

$ sgpt --code "write python code that converts 93 F in C" | tee /dev/tty | python
def fahrenheit_to_celsius(fahrenheit):
    celsius = (fahrenheit - 32) * 5/9
    return celsius

fahrenheit = 93
celsius = fahrenheit_to_celsius(fahrenheit)
print(celsius)
33.888888888888886

However, it'd be interesting to allow the model to call python iteratively—i.e. back and forth as many times as necessary.

[mdenesfe]

Hello,

Create a script that communicates with SGPT

  import subprocess
  
  def get_sgpt_response(query):
      sgpt_process = subprocess.Popen(
          ["sgpt", "--code", query],
          stdout=subprocess.PIPE,
          stderr=subprocess.PIPE,
          text=True,
          shell=True
      )
      stdout, stderr = sgpt_process.communicate()
      return stdout
  
  def run_python_code(code):
      python_process = subprocess.Popen(
          ["python"],
          stdin=subprocess.PIPE,
          stdout=subprocess.PIPE,
          stderr=subprocess.PIPE,
          text=True
      )
      stdout, stderr = python_process.communicate(input=code)
      return stdout
  
  while True:
      user_input = input("Enter your query: ")
      
      sgpt_response = get_sgpt_response(user_input)
      print(sgpt_response)
      
      python_code = input("Enter Python code: ")
      python_output = run_python_code(python_code)
      print(python_output)

and run sgpt_interactive.py