Bella is a digital companion application that combines voice interaction, visual expression, and AI dialogue to create a personalized conversational experience.
The project addresses the desire for a persistent digital presence by integrating voice recognition through Whisper ASR, video-based visual expression with cross-fading animations, and an LLM-powered dialogue system designed for natural conversation. The architecture centers on a singleton BellaAI class with modular components, running on Node.js with Express as the web service layer. Voice input flows through the speech recognition system, dialogue is processed through an optimized LLM with enhanced prompt engineering, and responses are presented through the web interface. The system manages AI model downloads automatically and supports microphone-based voice interaction through the Web Speech API.
Bella suits developers interested in exploring conversational AI interfaces and those wanting to experiment with multimodal interaction combining voice and visual feedback. The project is built with native JavaScript, HTML5, and CSS3 on the frontend, making it accessible for web-based development. The README does not compare Bella to alternative projects, so no comparative guidance can be offered. The application requires Node.js 22.16.0 or later and a modern browser with microphone permissions.
The project shows active development with implemented core features including voice perception, responsive UI design, and modular AI architecture. Several capabilities remain in the technology-ready phase, including speech synthesis and an emotional state system with basic infrastructure in place. The roadmap indicates planned expansion into memory systems, facial perception, multimodal interaction, and self-evolution mechanisms, suggesting ongoing architectural ambitions beyond the current feature set.
