Lynpoint/CyberVerse

Self hosted, real-time digital human agent platform. Build voice-first AI agents with WebRTC, persona memory, tools, RAG, and optional digital-human video.

What it solves

CyberVerse enables the creation of interactive digital humans that can see, hear, and speak in real time. It solves the complexity of coordinating low-latency voice interaction, persona-driven conversation, and synchronized video avatars, allowing users to bring a single photo to life as a living digital agent.

How it works

The framework uses a multi-agent architecture: a foreground PersonaAgent manages fluid, real-time conversations and interruptions, while background SubAgents handle long-running tasks like research or summarization. It utilizes WebRTC for low-latency communication and supports a plugin-based stack where components like the brain, voice, and face are replaceable. It can drive local models (like FlashHead or LiveAct) or use cloud-based digital human APIs.

Who it’s for

Developers and creators looking to build immersive AI agents, digital characters for entertainment, or interactive voice/video assistants that require persistent memory and specialized knowledge.

Highlights

  • Real-time video interaction: Supports full-duplex, natural video calls with digital humans.
  • Multi-agent architecture: Separates conversational flow from heavy computational tasks.
  • Character memory and RAG: Uses local storage for conversation history and RAG for knowledge-based persona alignment.
  • Extensible plugin system: Replaceable modules for LLMs, TTS, ASR, and vision via unified interfaces like LiteLLM.