WhatsLLM

A proof-of-concept integration between Twilio Whatsapp and OpenAI ChatGPT.

Description

Most conversational commerce companies are interested in how new AI developments may improve end user experience. This is a proof of concept project for using WhatsApp as an interface to OpenAI, allowing a customizable prompt and a short-term memory. The project can be modified to include some more elaborate features, such as using embeddings for a custom knowledge base, or integrating with other OpenAI features such as speech to text and image generation.

Getting Started

Prerequisites

• Python 3.10 - Check out Libraries And Frameworks
• Docker - Package the app for deployment
• SQL Database - Linked offering has a generous free tier
• Twilio Account - First users get free credits
• OpenAI API Key - First users get free credits
• Ngrok (or other network tunneling tool) for local development
• Fly.io CLI - Free tier up to 3 apps, easily package and run Docker based applications

Installing

• Clone this Repository

    git clone https://github.com/deistermatheus/whatsLLM.git

• Create an environment file, setting appropriate variables for Database, Twilio and OpenAI:

    cp .env.example .env

• Create the Database Structure

    alembic upgrade head

• Start the API

   uvicorn project.api.main:app --host 0.0.0.0 --reload

• Alternatively, to run a local docker image of the project:

    docker build -t whatsllm .
    docker run -it -p 8000:8000 --env-file=./.env whatsllm  bash server.sh

• Check the OpenAPI docs page or make a direct request to see the project is up and running:

   curl -X 'GET' \
  'http://localhost:8000/health' \
  -H 'accept: application/json'

• Setup your Twilio Webhook Endpoint

For local development, ngrok is recommended to get a public url for Twilio:

    ngrok http 8000

After obtaining a public URL, set up the Whatsapp Webhook in the Twilio Console. The docs for Twilio Webhooks can be found here.

Deploying

To start the deployment process, a Fly.io account is required:

    fly launch

The app requires a database connection and API keys for integrations, to set required environment variables:

    fly secrets import < .env

After making changes:

    fly deploy

The app should now be available in a public url such as:

    https://<your-chosen-app-name>.fly.dev/health
    https://<your-chosen-app-name>.fly.dev/docs

Deploying Automatically

Create a new secret in the repository with:

    fly tokens create deploy -x 99999h

Setup the GitHub Action by defining a new repository secret with the token

Expected outcome

• Chat with the bot over Whatsapp, the following capture is using the Twilio Sandbox Number:

Mini Roadmap

Features

• Get the integration working
• Add SQL Database for short term context
• Save customizable prompt in chatbot configs
• Deploy to the cloud
• Add Telegram as alternative channel
• Allow more customization choices over GPT (model, temperature...)
• Voice support
• Image generation
• Embeddings for larger context

Code Quality

• Add Automated Testing with CI Check
• Improve OpenAPI documentation
• Improve Type Hinting

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Explore Libraries and Frameworks

• FastAPI
• SQLAlchemy
• OpenAI SDK
• Twilio REST Client

Acknowledgments

• awesome-readme