Summary

Sakura is a universal desktop pet Agent framework designed for users seeking an interactive virtual companion. It allows importing custom characters that proactively perceive screen content and system events, reacting naturally to user scenarios, offering a more engaging and less demanding interaction than traditional chatbots.

How It Works

Sakura employs an agent engine that actively monitors the user's screen via periodic or on-demand screenshots, processing visual summaries to understand context. It integrates character packages (.char) defining dialogue, expressions, and voice, enabling proactive conversation initiation. The agent leverages a suite of built-in tools, including web search, reminders, and note-taking, to assist users based on detected scenarios, offering a dynamic and responsive virtual companion experience.

Quick Start & Requirements

The recommended installation is via pre-compiled releases, such as sakura-v0.9.8-windows-x64.zip for Windows. Installation involves unzipping the archive and running install.bat to set up dependencies, followed by start.bat to launch. Python 3.10+ is a core requirement. Mac and Linux users must consult the "完整安装指南" (Full Installation Guide). The "角色工作室" (Character Studio) for editing character packages is Windows-only. Official documentation links are available for setup, API configuration, and technical details.

Highlighted Details

• Customizable Characters: Supports importing .char packages containing dialogue, expressions, and GPT-SoVITS voice weights, with UI themes dynamically adapting to character aesthetics.
• Proactive Screen Awareness: Periodically captures screenshots to generate visual summaries for context, enabling characters to react to on-screen activities or user actions, even allowing region-specific screenshots.
• Rich Interaction: Features segmented bilingual replies, local quick responses during generation, typewriter text effects, and adaptive chat bubbles.
• Integrated Toolset: Includes built-in capabilities for screen capture, web browsing, to-do lists, reminders, note management, and long-term memory with hierarchical storage and recall.
• Flexible Voice Synthesis: Offers integrated GPT-SoVITS with hardware-tiered download options (RTX 50, Nvidia, CPU) and support for external GPT-SoVITS instances.
• Extensibility: Supports local plugins with automatic discovery and loading, and integrates with MCP Servers, including a built-in Web Search MCP.

Maintenance & Community

The project draws inspiration from the Shinsekai project and its ecosystem. While specific community channels (like Discord/Slack) or a public roadmap are not detailed in the README, the author is Rvosy. The project is actively maintained, as indicated by recent release versions.

Licensing & Compatibility

Sakura is released under the MIT License, permitting free use, modification, distribution, and sale, provided the original copyright and license text are retained. This license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The Windows platform is the primary focus for testing and development, with Mac and Linux users requiring specific setup procedures. The graphical "Character Studio" for package creation is exclusive to Windows. The GPT-SoVITS voice integration is tiered based on hardware capabilities, ranging from high-end GPUs to CPU-only configurations.