Why macvision?

Privacy by default

macvision sends nothing to a third party:

• Images are processed by Apple’s native Vision framework.
• No API keys, no image uploads, no cloud billing.
• The --screen option uses the local screencapture tool; the pixels never leave the Mac.

Why a CLI over the Vision framework?

Today, getting macOS vision capability into a script or an agent means one of three dead ends:

• shortcuts run — depends on a hand-built Shortcut, limited output, not scriptable.
• PyObjC / a Swift script — every agent re-wraps Vision, no shared interface, Python startup cost.
• sips — image conversion and metadata only; no vision.

macvision is the missing single binary: one JSON interface to the whole Vision surface, pipe-friendly, no per-user glue.

Why not Tesseract?

Tesseract is a capable cross-platform OCR engine, but:

• It is not Apple Vision — a different, generally less accurate model on macOS.
• It does OCR only — no scene classification, face/barcode/document detection, saliency, or feature-prints.
• It needs its own language data installed separately.

macvision uses the Vision framework that is already installed, optimized, and kept up to date by Apple, and it covers far more than text.

Why not cloud OCR / LLM vision?

• No network round-trip — images stay on the Mac.
• No API keys, no cost — uses the system framework.
• Predictable latency — no queue, no rate limits.
• Pipe-friendly — output JSON drives shell scripts and agents directly.

When you do want an LLM to reason about an image, pipe macvision ocr (or classify) output into it. macvision is the private, local sensing layer; the remote model is the reasoning layer.

When macvision is the right tool

• You need to read a screenshot, photo, or scan into text.
• You want to detect barcodes, faces, or document outlines in an image.
• You are building an agent that needs vision input on macOS.
• You want to compare or deduplicate images locally with feature-prints.