AI Agent Hub
Back to plugins
🖥️

dsh-quicksight

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Isanti2016/dsh-quicksight

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install Isanti2016/dsh-quicksight inside DeepSeek Harness to add the plugin; the source is available at https://github.com/Isanti2016/dsh-quicksight. After installation, restart the dsh web service and start a new session to use the quicksight_ocr tool.

About this plugin

A text-only model simply cannot see images, and that limitation blocks a lot of local workflows. dsh-quicksight plugs the gap with a single registered tool: call quicksight_ocr and get text or visual content back, with no extra orchestration and no manual routing required.

The engine is two-tier. Tier 1 runs RapidOCR entirely on local CPU, extracting Chinese and English text in seconds at zero API cost, with pixel data never leaving the machine. When OCR returns empty, too-short, or garbled output, or when the task truly needs visual understanding, it falls back to Tier 2, reusing the modlens ecosystem to call any configured visual engine: Nvidia NIM, Gemini, an OpenAI-compatible endpoint, or Anthropic. One tool call, best result, zero routing decisions for the user.

Ideal for developers running text-only models who want lightweight image recognition, and for privacy-conscious users who want to keep screenshots off third-party servers. The plugin listens on no ports, adds no runtime dependencies beyond Node built-ins, and opens no HTTP or WebSocket services. Clean to deploy, clean to remove.

Use Cases

  • A text-only model needs to extract text from a screenshot or document image
  • Understanding visual elements such as charts, layout, or colors in an image
  • Keeping sensitive images local without sending them to third-party APIs

Best For

  • Developers running text-only models who need lightweight image recognition
  • Privacy-conscious users who want to keep screenshots on their own machine
  • Engineers building local workflows that require an image understanding step