Fay is an open-source framework designed to connect digital humans (2.5D, 3D, mobile, PC, web) and large language models (OpenAI-compatible, DeepSeek) with business systems. It targets developers building interactive digital human applications, offering a flexible and modular architecture for integrating various AI models and deployment targets.

How It Works

Fay acts as a middleware, abstracting the complexities of different digital human models, LLMs, ASR, and TTS services. Its core design emphasizes modularity and interchangeability, allowing users to swap components like LLMs (e.g., OpenAI, DeepSeek) or voice services without significant code changes. This approach facilitates rapid prototyping and adaptation to diverse application requirements, from simple chatbots to sophisticated virtual presenters.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: Python 3.12, Windows, macOS, or Linux.
• Configuration: Modify ./system.conf.
• Run: python main.py or via Docker.
• Docs: https://qqk9ntwbcit.feishu.cn/wiki/UlbZwfAXgiKSquk52AkcibhHngg

Highlighted Details

• Fully open-source with a commercial-friendly license.
• Supports full offline operation.
• Offers real-time, stream-based processing.
• Enables flexible matching of digital human, LLM, ASR, and TTS models.
• Provides interfaces for text, voice, digital human driving, management, and automatic broadcasting.
• Supports custom knowledge bases, Q&A pairs, and persona information.
• Features wake-up and interruption capabilities.

Maintenance & Community

• The project has merged three previous versions into a single, more stable release.
• Community engagement is encouraged via a WeChat public account ("fay数字人").

Licensing & Compatibility

• License: MIT (stated as "商用免责" - commercial use exempt from liability).
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The project is actively evolving, with a note indicating a planned upgrade to the MCP protocol for agent decision-making. Some advanced integrations, like UE and Unity, are linked to separate repositories.