Miko
Embodied 3D AI desktop companion powered by Ollama, dual voice synthesis (Fish Audio S2.1 & local GPT-SoVITS), real-time lip-sync, and custom VRM avatar support
What is Miko?
Embodied 3D AI desktop companion powered by Ollama, dual voice synthesis (Fish Audio S2.1 & local GPT-SoVITS), real-time lip-sync, and custom VRM avatar support.
Overview of Miko
What it is
Miko is an embodied 3D AI desktop companion for Windows, powered by local large language models via Ollama. It integrates two voice synthesis engines—Fish Audio S2.1 Pro (cloud-based, low-latency) and GPT-SoVITS v2pro (local, offline)—and features a Unity-based borderless desktop avatar with real-time lip-sync and gesture animations. The project is open-source under GPL-3.0 and includes smart-home IoT control and optional hardware integration.
Key features
- Dual voice synthesis: Fish Audio S2.1 Pro (cloud WebSocket streaming) and GPT-SoVITS v2pro (local neural voice cloning).
- Real-time lip-sync and motion gestures via a Unity desktop avatar (Miko_desktop).
- Persistent conversation memory with temporal decay and vector importance scoring.
- Smart-home IoT device control via Tuya.
- Custom VRM avatar support (implied by avatar change option).
- Physical turntable hardware integration (OASIS) for face tracking.
Use cases
- Personal desktop AI companion for conversation and assistance.
- Offline voice interaction using local models and TTS for privacy.
- Smart-home control through voice commands.
- Customizable avatar for entertainment or productivity.
Getting started
Prerequisites: Windows 10/11, Python 3.10/3.11, Ollama installed. Pull a model: bash ollama pull qwen2.5:7b Run setup script: cmd setup.bat Configure .env with Fish Audio API key or local GPT-SoVITS paths. Launch with: cmd start_fishaudio.bat or cmd start.bat
License and pricing
License: GPL-3.0. Pricing: open source. No pricing details are stated in the README.
Based on the project's public README and repository metadata.