ArchiveGPT is an advanced, no-frills framework for deploying front-facing chatbots. It's like SillyTavern, but public-facing. The name comes from Blue Archive, because the first use of this framework was for a Blue Archive character. It is designed to be simple to deploy, and is heavily inspired by chikenuwu's Blue Archive chatbots MikaGPT and MiyakoGPT. His Github is here and I highly recommend you check out his work.
- It's customizable: ArchiveGPT is designed to be as plug-and-play as possible.
- It's lightweight: There's only 4 dependencies: Flask, Huggingface Transformers, ONNX Runtime, and NumPy.
- Python 3.x.x
- Flask
- Huggingface Transformers
- An AI provider of your choice (e.g. OpenAI, together.ai, SambaNova)
Clone the repository and install the required dependencies:
# Clone the repository
git clone https://github.com/mbhardt/ArchiveGPT.git
cd ArchiveGPT
# Install dependencies
pip install -r requirements.txt-
Set up your deployment:
- Set up
config.yaml - Write a system prompt for your character in
base_description.txt - Add the images of your character in
public/assets/sprites/{normal/alt/alt2} - Add the backgrounds you want in
public/assets/backgrounds - Change
SPRITE_COUNT,MAX_IMAGES, andSTUDENT_NAMEinpublic/assets/script.js(MAX_IMAGES refers to background images) - You're done!
- Set up
-
Run the server:
cd ArchiveGPT
python main.py- Access the chatbot:
Open your browser and go to
http://localhost:{port set in config.yaml}.
Contributions are welcome! Please fork the repository, create a new branch, and submit a pull request.
This project is licensed under the GNU General Public License v3.0 (GPL-3.0). See LICENSE for details.