Summary

Lanlan (Xiao8) is a user-friendly, audio-native AI companion designed for extreme low latency and full-scenario synchronization across devices. It targets users seeking an easily configurable, out-of-the-box AI partner, offering a lightweight and customizable experience without requiring a dedicated GPU.

How It Works

The project employs a Python backend focused on real-time multi-modal APIs, augmented by text-based plugins. The frontend utilizes H5+JS, packaged via Electron and PWA for cross-platform app deployment. Core design prioritizes minimizing voice interaction latency, ensuring seamless synchronization of a single AI companion's state and behavior across mobile, PC, and smart glasses.

Quick Start & Requirements

For ease of use, a one-click package with a launcher executable (新版启动器.exe) is available, requiring no GPU. Developers can clone the repository, set up a Python 3.11 environment, install dependencies (pip install -r requirements.txt), and configure API keys for third-party AI services (Alibaba Cloud or Zhipu recommended) in config/api.py. The core servers (memory_server.py, main_server.py) are then launched, accessible via a web interface (default: http://localhost:48911). API key configuration is mandatory for operation.

Highlighted Details

• Low Latency Focus: System architecture is optimized for minimal voice interaction delay.
• Cross-Device Synchronization: Enables a single AI instance to maintain consistent state and behavior across multiple devices simultaneously.
• Extensive Customization: Supports persona editing, Live2D model integration, and custom voice generation via short audio uploads.
• Flexible Model Support: Integrates with OpenAI Realtime API (supporting multiple providers like Zhipu, Alibaba) and open-source models including Qwen-Omni-Realtime and GPT-Realtime.
• Agent Mode & Memory Management: Features an Agent mode for advanced control and a memory browser for reviewing and correcting AI recall.

Maintenance & Community

The project is actively developed, with recent updates including Agent mode and new model support. Community interaction is facilitated via a QQ group (1048307485). Demo videos are available on Bilibili.

Licensing & Compatibility

Distributed under the MIT license, offering broad compatibility for commercial use and integration into closed-source projects.

Limitations & Caveats

The project does not guarantee AI intelligence levels, focusing instead on performance optimization. Live2D mouth synchronization issues are typically model-dependent. Tool calling functionality is limited on certain platforms (e.g., Alibaba), and real-time model integration requires careful handling of asynchronous operations. Some model integrations (e.g., Gemini Live) are geographically restricted.