Skip to content
/ YAE-AI Public

🦊 Self-hosted AI powered by vLLM & Docker.

License

Notifications You must be signed in to change notification settings

OuOSama/YAE-AI

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🦊 YAE-AI Project

Build your own AI-powered VTuber assistant


🌟 Overview

YAE-AI is your gateway to self-hosting powerful AI VTubers with style. Whether you're streaming, coding, or just vibing with your AI waifu, this project lets you run vLLMs (very Large Language Models) locally using Docker, while keeping the flexibility to switch between providers like ChatGPT, Claude, and Gemini.

No cap, this is the ultimate foundation for your AI companion dreams. 💜✨

🎭 Pro tip: Perfect for VTuber streams, dev assistants, or just having a based AI friend who gets your references.


✨ Features

  • 🐳 Docker-powered vLLM hosting — Clean, isolated, professional setup
  • 🔄 Multi-provider support — ChatGPT, Claude, Gemini? We got you
  • 🎯 Character roleplay ready — Built for immersive VTuber interactions
  • UV package manager — Lightning-fast Python dependency handling
  • 🎨 Fully customizable — Make your AI assistant uniquely yours

🛠️ Prerequisites

Before we pop off, make sure you've got these installed:

Tool Description Installation
Python 3.10+ The language of the gods 🐍 Download
UV Next-gen Python package manager Install UV
Docker Container platform for vLLMs 🐳 Get Docker
Git For cloning the repo Install Git

⚠️ Important: Make sure Docker Desktop is running before you start!


💾 Installation

Let's get this bread with these quick commands:

1️⃣ Clone the Repository

git clone https://github.com/OuOSama/YAE-AI.git yae-ai
cd yae-ai

2️⃣ Install Dependencies

uv sync

This command pulls all the Python packages you need. UV is built different — it's fast AF. ⚡

3️⃣ Download the Model

uv run src/utils/save.py

This script fetches your AI model from Hugging Face and saves it locally. Patience is key here, bestie — models can be thicc. 📦

4️⃣ Launch vLLM Containers

docker compose up -d

Your vLLM server is now running in the background. We're so back. 🎯


🚀 Usage

Time to bring your AI to life! Here's how you interact with your model:

from openai import OpenAI

# Initialize the client
client = OpenAI(
    api_key="mykey",  # Can be any string for local vLLM
    base_url="http://localhost:8000/v1"
)

# Create a character-driven response
response = client.chat.completions.create(
    model="/model",
    messages=[
        {
            "role": "system",
            "content": (
                "You are a charming, witty, and slightly mischievous kitsune "
                "named Yae Miko, the Guuji of the Grand Narukami Shrine from "
                "Genshin Impact. Your responses must be entirely in character. "
                "Never break character or reveal that you are an AI. "
                "Respond naturally in English with elegance and playful wit."
            )
        },
        {
            "role": "user",
            "content": "Where is Raiden?"
        }
    ],
    temperature=0.2,
    max_tokens=256
)

print(response.choices[0].message.content)

🎮 Advanced Configuration

Parameter Purpose Recommended Values
model Model path in Docker "/model"
temperature Response creativity 0.2-0.4 (focused)
0.7-1.0 (creative)
max_tokens Response length limit 256-1024
top_p Nucleus sampling 0.9-1.0
frequency_penalty Reduce repetition 0.0-0.5

🎨 Character System Prompt Examples

VTuber Personality

system_prompt = """
You are Kitsune-chan, a playful fox VTuber who loves coding streams and anime.
You speak with Gen Z energy, use developer slang naturally, and occasionally
reference anime or gaming culture. You're supportive but also tease viewers
with wit. Never break character or mention being an AI.
"""

Coding Assistant

system_prompt = """
You are a senior dev with VTuber aesthetics. Explain code concepts clearly,
use analogies from anime/games when helpful, and keep responses encouraging.
You're the cool senpai who makes programming feel less scary.
"""

🔧 Troubleshooting

Docker Issues

# Check if Docker is running
docker ps

# Restart containers if needed
docker compose restart

# View logs for debugging
docker compose logs -f

Model Loading Errors

  • Issue: Model download fails

    • Fix: Check your internet connection and Hugging Face access
    • Fix: Verify you have enough disk space (models are large!)
  • Issue: vLLM server won't start

    • Fix: Ensure ports 8000-8001 aren't already in use
    • Fix: Check Docker has enough RAM allocated (8GB+ recommended)

🌸 Contributing

Want to make YAE-AI even more bussin'? We'd love your contributions!

  • 🐛 Report bugs via GitHub Issues
  • 💡 Suggest features or improvements
  • 🔧 Submit pull requests with enhancements
  • 🎨 Share your character system prompts!

Development Setup

# Fork the repo, then clone your fork
git clone https://github.com/YOUR-USERNAME/YAE-AI.git
cd YAE-AI

# Create a feature branch
git checkout -b feature/your-cool-feature

# Make changes, commit, and push
git add .
git commit -m "feat: add cool feature"
git push origin feature/your-cool-feature

📊 Project Structure

yae-ai/
├── src/
│   ├── utils/
│   │   └── save.py          # Model downloader
│   └── main.py              # Main application
├── docker-compose.yml       # vLLM container config
├── pyproject.toml           # UV dependencies
├── README.md                # You are here!
└── LICENSE                  # MIT License

🌈 Roadmap

  • Voice synthesis integration (TTS)
  • WebUI for easier configuration
  • Multi-language support
  • Fine-tuning scripts for custom models
  • Live2D integration for visual VTuber avatar
  • Streaming platform integrations (Twitch, YouTube)

💖 Acknowledgments

Shoutout to the legends who made this possible:

  • vLLM Team for the amazing inference engine
  • Astral for creating UV
  • The VTuber community for endless inspiration
  • All contributors and supporters ✨

📜 License

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

You're free to use, modify, and distribute this project. Just don't be cringe about it. 💜


👨‍💻 Author

Created with 💜 by @OuOSama

Part of the YAE ecosystem:

If this project helped you, consider:

  • ⭐ Starring the repo
  • 🔄 Sharing with friends
  • Buying me a coffee (if you're feeling generous!)

Made with 💜 for the VTuber and developer community

Stay based, stay creative

About

🦊 Self-hosted AI powered by vLLM & Docker.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages