Harzva/dsh-maclens

README 摘要

dsh-maclens 🍎🔍 Apple's on-device Vision framework , bridged into DeepSeek Harness ( dsh ) as local tools: OCR, image classification, face detection, document layout, and a combined read — 100% offline, no API key, no daemon . 🔒 Privacy Every pixel stays on your Mac. No network, no upload, no telemetry. ⚡ Speed Sub-second OCR on typical screenshots (Neural Engine). 🇨🇳 Languages zh-Hans + 30+ recognition languages out of the box. 🖼️ Tall images slice splits long screenshots so small text survives Vision's downscale. 🧩 No deps Swift CLI ships in the npm package — no build step to install. 👤 For humans — quick start (30 seconds) The five tools the model can call: Tool One-liner maclens ocr "Read all the text in this image" — every line + confidence + box maclens classify "What kind of image is this?" — document, chart, photo, … maclens faces "Are there people in this image?" — face boxes + count maclens document "Parse this page" — OCR + left/right column layout maclens describe "Give me everything at once" — OCR + classify + faces + layout When to pick maclens vs a VLM: maclens is a CV toolkit — it transcribes, classifies, detects but does not narrate "what this image is about…

分类