fahsq72lan/dsh-guide-tutor

guide-tutor consists of two parts: the DeepSeek Harness plugin and a native Windows overlay. Clicking the pinned floating ball captures the screen and sends it to the current conversation. After the model analyzes the image, it draws rectangles, circles, arrows, and text on a full-screen transparent layer, while a semi-transparent panel displays the current step. Responses are read aloud sentence by sentence using local Chinese text-to-speech with subtitles, and voice input is recognized locally offline using sherpa-onnx. Tutorial steps can be saved to a local Markdown or Word knowledge base for direct reuse next time.

guide-tutor consists of two parts: the DeepSeek Harness plugin and a native Windows overlay. Clicking the pinned floating ball captures the screen and sends it to the current conversation. After the model analyzes the image, it draws rectangles, circles, arrows, and text on a full-screen transparent layer, while a semi-transparent panel displays the current step. Responses are read aloud sentence by sentence using local Chinese text-to-speech with subtitles, and voice input is recognized locally offline using sherpa-onnx. Tutorial steps can be saved to a local Markdown or Word knowledge base for direct reuse next time.

Why It Matters It transforms "explaining steps based on a screenshot" into direct on-screen annotations with voice narration, making it more intuitive when teaching older relatives or new users how to use software.

Star count 62

Reasons

  • The repository 11 was created on 13, and after reading the README, the features and data storage locations are clearly documented
  • Text-to-speech and speech recognition are both performed locally, requiring no additional API keys

Usage

  • In DSH, use plugin_manager's install_bundle to point to the plugin directory for installation, then describe the operation you want to learn in the panel. Once the model confirms the plan, it begins annotating

Risks

  • Supports Windows only; screen reading sends a full-screen screenshot to the model being used, so sensitive windows must be closed first
  • The repository does not declare a license

Open

GitHub

Explore the latest AI Radar updates