Fay is an open-source agent framework that connects large language models to digital humans and real business systems. It lets a virtual avatar in 2.5D, 3D, mobile, PC or web form hold natural voice conversations, powered by the LLM of your choice, and it has collected roughly 13,600 stars on GitHub while remaining in active development through 2026.
Most LLM demos stop at a chat box. Fay is built for the step beyond that: a digital human that can speak, move its lips in sync with its voice and act as the front end of a business process. The framework plugs into OpenAI-compatible APIs and models such as DeepSeek, combines speech recognition, language understanding and speech synthesis into one loop, and uses Wav2Lip for realistic lip synchronization. On the presentation side it integrates with Unity3D, Unreal Engine 5 and Android, which makes it practical for kiosks, live-streaming rooms and in-app assistants. Fay also exposes interfaces for connecting the avatar to external business systems, so the digital human can query orders, trigger workflows or act as a live-commerce host rather than just chatting.
Fay is completely free and open source under the GPL-3.0 license. You run it on your own machines and only pay for any third-party LLM or speech API keys you choose to plug in.
Fay is for developers and solution builders creating digital-human customer service agents, virtual live-stream hosts and interactive kiosk assistants, particularly in Chinese business scenarios where an on-screen avatar needs to converse and transact, not just answer questions.