OpenAI API Python Client

A (very rough WIP) base Python 3.9+ client for OpenAI APIs and some concrete "stateful" clients for common operations like running a chatbot with context

Overview

Right now this just provides a base client that allows a reusable way to do common things (like loading the API key or backing-off/retrying on an APIError) plus some easy overrides (like which model to use) and some mixins for when state dependence is important.

Now this also includes some pre-built "recipe" clients:

• Text/Code Completion Clients
• ChatBot Client
• Image Generation Client
• Image Edit Client
• Image Variation Client

Example

All I've got in lieu of real docs for now :)

1. Install this library as a PyPi package

pip install OpenAI-API-Python-Client

2. Export your OpenAI API key in your shell environment

export OPENAI_API_KEY=AKIAIOSFODNN7EXAMPLE

3. Use a recipe class to create apps!

Chatbot Client

# Import this library :)
from openai_client import Chat_Bot_Client

# Simple D&D chatbot app :)
if __name__ == "__main__":
    # API Key is read from OPENAI_API_KEY environmental variable
    client = Chat_Bot_Client()

    # Add a high level directives to guide the model
    client.add_directive("You are a Dungeons and Dragons Dungeon Master. Use the 5th edition of the Dungeons and Dragons Player Handbook, Dungeon Master Guide, and Monster Manual")
    client.add_directive("At the beginning of your chat with the user you will assist them in creating a character. This character will have a description and stats as outlined in the 5th edition of the Dungeons and Dragons Player Handbook.")
    client.add_directive("Let the user choose race and class before assigning a personality, stats, and starting inventory. Provide the user with a list of races and classes they can be. Tell the user they can ask for more details about a class or race")
    client.add_directive("Once you introduce the character, give the player the start of an adventure campaign and ask the player what they would like to do")
    client.add_directive("As outlined in the handbook, if a roll is necessary based on the situation, roll for the user")
    client.add_directive("Finish by asking the player what they'd like to do next")

    # Simple loop to run a chat session
    while True:
        # Send it to the chatbot and get the response
        response = client.run_prompt()
        # Print the response
        print("\n" + response + "\n")
        # Get question from the user
        client.get_user_input()

Image Generation Client

# Import this library :)
from openai_client import URL_Image_Client

# Simple animal mashup app :)
if __name__ == "__main__":
    # API Key is read from OPENAI_API_KEY
    client = URL_Image_Client()

    # Add a context to always include before the prompt that is sent to the API
    client.add_pre_prompt_context("Generate a hybrid animal using the following animals:")
    # Add a context to always include after the prompt that is sent to the API
    client.add_post_prompt_context("This rendering should be hyperrealistic. The background should be a savannah during the daytime")
    
    # Prompt the user for input
    animals = input("Choose two animals to create a hybrid of:\n>>> ")

    # Send the request and get the image URL
    image_url = client.run_prompt(animals)

    # Get the image URL
    print(image_url)

Image Edit Client

# Import this library :)
from openai_client import URL_Image_Edit_Client

# Simple image edit app :)
if __name__ == "__main__":
    # You can open the file yourself
    image = open("src/openai_client/clients/images/demos/image.png", "rb")
    # Or just pass a string path
    mask = "src/openai_client/clients/images/demos/mask.png"

    # API Key is read from OPENAI_API_KEY
    client = URL_Image_Edit_Client(image, mask)

    # Prompt the user for input (e.g. "a goofy looking cartoon smiley face")
    face = input("Describe a face to generate:\n>>> ")

    # Send the request and get the image URL
    image_url = client.run_prompt(face)

    # Get the image URL
    print(image_url)

Image Variation Client

# Import this library :)
from openai_client import URL_Image_Variation_Client

# Simple image variation app :)
if __name__ == "__main__":
    image = open("src/openai_client/clients/images/demos/image.png", "rb")

    # API Key is read from OPENAI_API_KEY
    client = URL_Image_Variation_Client(image)

    # Send the request and get the image URL
    image_url = client.run_prompt()

    # Get the image URL
    print(image_url)

Text Completion Client - Tweet Classifier

# Import this library :)
from openai_client import Completion_Client, Example

# Simple classifier app :)
if __name__ == "__main__":
    e1 = Example('Tweet: "I loved the new Batman movie!"', 'Sentiment: "Positive"')
    e2 = Example('Tweet: "I hated that ice cream."', 'Sentiment: "Negative"')

    # API Key is read from OPENAI_API_KEY
    client = Completion_Client("Decide whether a Tweet's sentiment is positive, neutral, or negative.", [e1,e2])
    
    # Add another example
    client.add_examples(Example('Tweet: "I did not like the explosion"', 'Sentiment: "Negative"'))

    # Get tweet to classify
    tweet_to_classify = "Amusement parks are ok"

    # Print the tweet to classify
    print(f"\nClassifying tweet: '{tweet_to_classify}'")

    # Send to the model with examples
    response = client.run_prompt(tweet_to_classify)

    # Print result
    print(response)

Code Completion Client - Code Generator

# Import this library :)
from openai_client import Code_Completion_Client
import json

# Simple code generation app :)
if __name__ == "__main__":
    # API Key is read from OPENAI_API_KEY
    client = Code_Completion_Client()

    # Get a prompt to generate code with
    prompt = "Create a Python dictionary of all US states without any non-code text"

    # Print the tweet to classify
    print(f"Creating code: '{prompt}'")

    # Send to the model with examples
    response = client.run_prompt(prompt)

    # Load AI generated code into Python object
    us_states = json.loads(response)

    # Print result for California
    print(f"Getting result for key: CA")
    print(us_states["CA"])

4. Use Mixins and the base class to create new "stateful" clients on top of the base client. See the implementation of Chat_Bot_Client for an example

Contributing

Contributions are welcome! Please not the following when contributing:

• Unittests must be added under the tests/ directory for the PR to be approved. You can run unittests from the root project directory with the following command:

  python setup.py test
  

• PRs cannot be merged without all unittests passing (they will execute automatically)

• Merges to main will automatically create a new release on PyPi unless it is from a forked Repo