Developer Beni launched a web app for real-time video calls with an AI companion, moving beyond text chatbots. The core is a Live2D avatar that animates with expressions and motion during conversation, driven by real-time voice streaming and a long-term memory system for continuity across sessions.

The main technical hurdle was synchronizing the entire loop: microphone input, model response, TTS audio, and 2D animation. The developer spent more time on state management, latency, and buffering than on prompts, emphasizing the challenge of making an AI feel present and responsive in a live call.

Currently, sign-in is required to persist memory and prevent abuse, though a guest mode is planned. The project is browser-based with client-side playback control and hooks for speech-driven animation. It joins a growing trend of AI companions aiming for more immersive, human-like interaction beyond simple text exchanges.