youtu-tip  by TencentCloudADP

On-device AI assistant for intelligent work understanding

Created 3 weeks ago

New!

541 stars

Top 58.9% on SourcePulse

GitHubView on GitHub
Project Summary

Youtu-Tip is a proactive, on-device AI assistant that intelligently understands users' current work, offering a privacy-preserving alternative to cloud solutions. It targets users valuing data security and reduced AI interaction, providing a seamless "AI super entry" experience. Benefits include enhanced privacy and productivity via intelligent context detection and local agent execution.

How It Works

Tip extends Youtu-Agent, integrating agent invocation and contextual intent detection. Its core is a minimal-interaction "AI super entry" activated by a hotkey. Users select text or capture screenshots; Tip prepares the context. It's powered by self-developed lightweight models: Youtu-LLM (1.96B parameters) with native agent capabilities, and Youtu-VL (4B multimodal, soon open-source). This architecture enables fully offline, on-device processing for privacy and local agent task execution.

Quick Start & Requirements

Installation is via GitHub Releases. Support is for MacOS with Apple Silicon (M-series). Screen recording and accessibility permissions are required. Activation is via ctrl + shift hotkey, supporting direct chat, context-aware chat (text selection), or screenshot interaction. Tip connects to on-device models via Ollama or OpenAI SDK-compatible endpoints. Official quick-start guides and documentation links are available: Website, Tip Overview, Using Tip, Quick Start Guide.

Highlighted Details

  • AI Super Entry: A single hotkey (ctrl + shift) provides a unified AI interface, minimizing user input via screen context.
  • On-Device Privacy: All data processing occurs locally, supporting fully offline model calls for enhanced security.
  • Youtu Agent Integration: Enables local execution of complex tasks, including file management and tool interaction, via configurable agent environments.
  • Youtu-LLM: A compact 1.96B parameter model demonstrating SOTA performance in STEM, coding, and agent benchmarks, outperforming similarly sized models.

Maintenance & Community

No specific community channels (e.g., Discord, Slack) or detailed roadmap information are provided in the README.

Licensing & Compatibility

The project is open-sourced under a specified LICENSE file. Compatibility is currently limited to MacOS devices with Apple Silicon.

Limitations & Caveats

The application is restricted to MacOS with Apple Silicon hardware. The Youtu-VL multimodal model is not yet open-source.

Health Check
Last Commit

1 week ago

Responsiveness

Inactive

Pull Requests (30d)
8
Issues (30d)
16
Star History
543 stars in the last 27 days

Explore Similar Projects

Feedback? Help us improve.