v0.1.0

First release.

A DeepSeek Harness plugin that reads text out of screenshots, clipboard images, and image files using macOS's built-in Vision OCR. No vision model, no API key, no upload. The image is processed on your Mac.

Install

dsh plugin --profile web add github:uknowmyface/locallens

The package ships its own cordis.patch.yml via dsh.bundle.patch, so install and registration happen in one step. Then ask in the chat box: "OCR the clipboard" / "看看剪贴板这个报错".

Requires macOS 13+ and Xcode Command Line Tools (xcode-select --install).

Before you trust it

• The image stays local, but the recognized text is a tool result: it enters the model context and goes to your model provider like anything else you type. See Privacy boundary.
• OCR output is untrusted input. Results are wrapped in <untrusted-ocr-text> and the model is told to treat them as data, but that is a prompt-level mitigation, not a boundary. See Security notes.
• source=screen captures the entire main display.
• The code was written by AI and has had no line-by-line human review. ~800 lines, no runtime dependencies. Please read it.

Contributions wanted

A Windows port is the biggest gap: Windows.Media.Ocr is structurally the same as Apple Vision, and only the engine layer behind lib/ocr-engine.js needs replacing. Also wanted: region and window capture, a Linux port, PDF input, language accuracy testing.