N.E.K.O: a proactive AI companion platform with persistent memory and multi-avatar support

N.E.K.O: a proactive AI companion platform with persistent memory and multi-avatar support

What it solves

N.E.K.O. is designed to move beyond simple task-oriented AI agents or static roleplay chat interfaces. It aims to create a "digital life"—an AI companion that possesses real-time awareness of the user's environment, persistent memory, and the ability to proactively initiate interaction and assist the user within their digital workspace.

How it works

The project integrates several advanced AI capabilities into a single platform:

  • Perception: Uses Realtime APIs for voice and ChatCompletion for text, combined with visual understanding to "see" the user's screen.
  • Memory: Employs a five-dimensional memory system (working, recent, factual, reflective, and persona memory) to ensure the companion evolves and remembers the user over time.
  • Interaction: Supports multiple avatar formats (Live2D, VRM, MMD, PNGTuber) with motion capture and full-screen tracking.
  • Agentic Capabilities: Can operate browsers and computers via tools like CUA and OpenClaw A2A to perform actual tasks.
  • Extensibility: Features an SDK and plugin store for scenarios like gaming companionship, social media interaction, and smart home integration.

Who it’s for

  • Users seeking an emotionally intelligent, proactive AI companion rather than a tool.
  • Creators (artists, 3D modelers, voice actors) who want to build and share custom AI characters via the Steam Workshop.
  • Developers interested in end-to-end integrated AI companion frameworks.

Highlights

  • Proactive Engagement: The AI can initiate conversations based on screen content, trending topics, or personal dynamics.
  • Cross-Platform Memory: Syncs personality and memory across desktop, mobile, and games.
  • ** uma Open-Source Core**: The core driver is licensed under Apache 2.0 and runs locally to ensure data privacy.
  • Multi-Model Support: Compatible with 14+ AI providers including OpenAI, Gemini, Qwen, and DeepSeek.

Sources