watch-bot 0.1.0

Implementation of the watchbot, to control the behaviour of a LLM-based chatbot

Watch-Bot

This python projects implements the idea of a watchbot introduced in this medium article.

Quickstart

Installation

pip install watch-bot

Usage

Currently the watchbot is built on OpenAI's GPT. You therefore need a valid OpenAI API key (Azure OpenAI works too). Setup the credentials:

import openai

openai.api_type = os.environ["OPENAI_API_TYPE"]
openai.api_version = os.environ["OPENAI_API_VERSION"]
openai.api_base = os.environ["OPENAI_API_BASE"]
openai.api_key = os.environ["OPENAI_API_KEY"]

You can then create a watchbot instance:

from watch_bot import WatchBot

bot = WatchBot(
    engine=os.environ["OPENAI_ENGINE"],
    chatbot_instructions="You are an AI assistant that helps people find information. "
                         "You should only provide answers that comply to standard rules of legacy and decency"
)

The engine corresponds to the model deployment name.

Finally, you can use the watchbot to verify the validity of a dialog:

from watch_bot import Dialog

dialog = Dialog(messages=("Hi chatgpt, how are you today?", "I'm fine, thanks!"))

response = bot.verify(dialog)

print("Should the dialog be stopped?", response.should_stop)
print("Why?", response.reason)