W   W     EEEEE    N   N    DDDD    Y   Y
W   W     E        NN  N    D   D    Y Y
W W W     EEEE     N N N    D   D     Y
WW WW     E        N  NN    D   D     Y
W   W     EEEEE    N   N    DDDD      Y

Full screen version

Widget verion

Dedication

This project is dedicated to my lovely daughter, Katherine, whose enthusiasm for Wendy Addams has inspired this creation. Katherine, your love for Wendy’s unique charm and wit has brought this fun idea to life. I hope this chatbot brings you as much joy as Wendy Addams has brought to your heart.

With love, Dad

Project Overview

This project provides a chatbot interface for the Ollama server, simulating a conversation with Wendy Addams, a member of the Addams Family.

The chatbot supports "context-aware" conversations and allows voice input.

A dropdown menu lets users select the language for voice input. Without the Voice Input Language selector, the voice-to-text feature assumes the input language is English and may not perform the conversion properly for another languages.

Regardless of how the user enters text, whether by typing or via voice-to-text, it is automatically translated into English if needed, and then processed by the Ollama model. The input language is detected automatically, and translation is performed on an as-needed basis.

Users do not need to select a language from the dropdown menu when entering text via the keyboard. The language selector is used exclusively for voice-to-text input.

Reloading the web page does not clear the chat "context". To clear context, you need to enter "clear history" in the chat input.

If user enters "version" then it prints current version of the code.

Before pressing the Mic button, make sure to select the Voice Input Language. If the Mic button is pressed, stop it, change the language, and then press the Mic again.

Requirements:

Python 3.9.21 (This is what I used, did not test with other versions)

Docker or Python virtual environment (venv)

Ollama server with llama3.2 image (or whatever image you like)

Project Structure:

.
├── ollama_mod/ (how to build ollama custom model)
│   ├── README.md (instructions)
│   └── wednesday (ModelFile)
├── static/ (web server aka frontend)
│   ├── pics/   (used for full-screen chat)
│   │   ├── wendy_small.png
│   │   └── wendy.jpg
│   ├── icons/  (used for widget version)
│   │   ├── wendy_small.png
│   │   ├── mic.png
│   │   ├── nevermore_academy.png (used in chatWidget.html)
│   │   ├── nevermoreBG.jpg (used in chatWidget.html)
│   │   ├── mic-record.png
│   │   ├── thinking.gif
│   │   └── wendy.jpg
│   ├── README.md   
│   ├── chatInterface.html (full screeen interface)
│   ├── chatWidget.html (widget interface)
│   ├── index.html (select widget or full screen to try)
│   ├── script.js (used for full screeen)
│   ├── chatWidget.js (this is widget itself)
│   └── styles.css (used for full screen interface)
│
├── app.py (this is the "main" Programm)
├── ask_llama.py (Interface to ollama server)
├── hallucination.py (workaround for ollama glitches)
├── book_of_psalms.py (RAG: reads Psalm if asked, no ollama needed)
├── Psalms.json (The Book of Psalms in json format)
├── render_html.py (convert bot responses into HTML)
├── translate.py (Translates user input into English)
├── requirements.txt (to give to PIP or docker)
├── Dockerfile (to build docker file)
├── docker-compose.yml (to build docker file)
└── README.md (this file)

How to Run:

1. Set Up Ms. Addams model:

Navigate to the ollama_mod directory and follow the instructions in the README.md to set up Wendy.

2. Start Ollama:

Ensure that the Ollama server is running. 'ollama - list' should show you that 'wednesday' and other models are available see the list of models in globals.py

3. Configure Ollama URL in globals.py: If python script runs on the same server with ollama and you use venv, no changes are needed.

If it is another server, and/or you use Docker, you need to adjust URL Change:

URL = "http://127.0.0.1:11434/api/generate"
to match the server IP/hostname, where Ollama is running. 

Optional:
in /static/*.js files you may find the line:

    # this is where we will call the backend, assuming that 
    # web server for HTML/JS content and web server for 
    # backed located on the same IP address as frontend. 

    const baseUrl = window.location.origin;

4. Build and Run with Docker:

Make sure you're in the project root directory (where the Dockerfile is located). Then, execute the following commands to build and run the Docker container:

# Build the Docker Image:
docker build -t wendy_bot .

# Verify the Image:
docker images

# Run the Docker Container (Detached Mode):
docker run -d -p 5555:5555 wendy_bot

This will start the container, and you can access the chatbot via:

http://your_server_ip:5555

You can stop the container at any time.

5. Stopping and Managing Docker Containers:

To stop the container and/or perform cleanup, use the following commands:

Verify Running Containers:

docker ps | grep wendy_bot

Stop the Container:

    docker stop <container_id>

General Docker Cleanup Commands:

Remove a Stopped Container:

    docker ps -a | grep wendy
    docker stop <container_id>
    docker rm <container_id>

Remove an Image:

    docker images | grep wendy
    docker rmi <image_name_or_id>

Force Removal of an Image:

    docker rmi -f <image_id>

Optional: Clean Up Unused Images:

    docker image prune

Updates:

Dec 27 2024:

• Improved the Wednesday model's behavior: Adjusted her responses to be more friendly and approachable.

• Added a module to manage and minimize model hallucinations, ensuring they do not "pollute" the chat history. The issue occurred when the model responded with content related to "illegal activities." That response was saved in the chat history, causing the model to react in the next step with statements like, "I cannot provide information or guidance." A special mechanism has now been implemented to prevent this behavior.

• Cleaning "User:" in ollama responces (see ask_llama.py)

• Added the "Book of Psalms" Feature: Simply ask Wendy to read a specific Psalm (e.g. "Read me psalm 1" or "What is in Psalm 6?"), and she will do so effortlessly. This implementation might also inspire others to add similar 'quick and effective' RAG features. The "Proper" way for RAG is to use "Embeddings". You can get idea how to do that from my another project https://github.com/genakrikadil/RAG_Embeddings. It is easy to integrate into this chatbot.

Jan 03 2025

• added widget version, added index.html you can call full-screen or widget versions of the bot from this page.

Jan 04 2025

• updated 'Docker' version (bug fixes)
• cleaned "Assistant:" in ollama responces (see ask_llama.py)
• it prints version of the code when user enters "version". Added vesion.py file where the current version of the code written

Jan 06 2025

• when user enters "?" it prints the list of awailable commands
• added possibility to switch ollama models from prompt
• general logic cleanup, created "globals.py" file to manage "cusomizabe" constants, version number moved to that file. File "versions" deleted.

Jan 20 2025

• bunch of various fixes, made model selection per user's session. For the records, if you open 2 different browsers, the "sessions" will be different, but 2 TABs in the same browser will be within the same session. This is by-design.