Jaison-core provides a configurable response server for Project J.A.I.son, enabling users to create dynamic and interactive AI-driven applications. It targets developers and enthusiasts looking to build custom AI assistants, particularly those integrating with platforms like Twitch and VTube Studio, by offering a flexible backend for managing AI responses and component interactions.

How It Works

Jaison-core acts as a central hub, orchestrating communication between various AI components (Speech-to-Text, Text-to-Text, Text-to-Speech) and applications. It leverages a plugin system, allowing users to specify and configure different AI models and services, including local options like OpenAI Whisper and Unsloth, or API-based services. This modular design facilitates customization and integration with a wide range of AI technologies and user-facing applications.

Quick Start & Requirements

• Installation: Conda virtual environment recommended.

  conda create -n jaison-core python=3.12 ffmpeg cudatoolkit -c pytorch -y
  conda activate jaison-core
  pip install -r requirements.txt
  

• Prerequisites: Python 3.12, Conda, FFmpeg, CUDA, PyTorch (machine-specific install), Nvidia GPU (RTX 4070 mentioned, non-Nvidia likely requires code stripping).
• Setup: Requires setting up and linking external STT, T2T, TTSG, and TTSC components (local or API-based).
• Running: python ./src/main.py --config=config.json
• Docs: J.A.I.son Setup

Highlighted Details

• Supports local AI models (e.g., OpenAI Whisper, Unsloth) and API-based services.
• Plugin architecture for modular component integration (STT, T2T, TTSG, TTSC).
• Configuration options for prompts, translations, and conversation history.
• Detailed logging for system, dialog, and AI response training.

Maintenance & Community

• Community-driven plugins are available via Discord.

Licensing & Compatibility

• The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

The project is explicitly stated to be unlikely to run on non-Nvidia systems without code modification, and requires specific AI components (STT, T2T, TTSG, TTSC) to be functional. The current plugin system allows only one of each component type.