Pangram verdict · v3.3
We believe that this document is primarily human-written, with some AI-generated content detected
AI likelihood · overall
MixedArticle text · 1,407 words · 9 segments analyzed
v1.1.0 · 100% on-device · MIT Apple Vision from your terminal.Point auge at any image. Get OCR text, classification labels, barcode payloads, and face bounding boxes back. One command, every analysis. The Vision framework is already on your Mac - auge gives you a UNIX CLI to it. 100% on-device No API keys Zero dependencies Pipe-friendly$brew install Arthur-Ficial/tap/auge Source: github.com/Arthur-Ficial/augeRequires: macOS 10.15+ · Apple Silicon or Intel · No Xcode needed Everything runs on your device. No network calls. No data leaves your Mac. Enforced at runtime by a URLProtocol guard that hard-exits any HTTP attempt. Terminal — auge$ auge --all photo.jpg === OCR === (no text) === CLASSIFY === animal: 92% cat: 92% feline: 92% mammal: 92% === BARCODES === (none) === FACES === 0 faces detected $ auge --all -o json scan.pdf # PDF natively, full structured output $ auge --all --md doc.png | apfel "summarize" $ auge --all --clipboard # NSPasteboard image input $0Per requestNo API keys, no usage tier, no per-token billing. Run it a million times.100%On-deviceZero network calls. Hard-enforced at runtime - any HTTP attempt aborts.0DependenciesPure Swift, no third-party packages. Vision ships with macOS.187Tests passingPure-Swift test runner, no XCTest. Same pattern as apfel. OCRauge --ocr image.pngText recognition via VNRecognizeTextRequest. Handles screenshots, scans, photos of signs, even handwriting. PDF input goes through PDFKit (text layer when present, rasterize for OCR otherwise).$ auge --ocr --langs en-US,de-DE doc.png Hello World Grüß GottClassifyauge --classify image.jpgImage classification via VNClassifyImageRequest. 1000+ categories with confidence percentages.
Animals, vehicles, objects, scenes, materials - the same model Photos uses.$ auge --classify cat.jpg animal: 92% cat: 92% feline: 92%Barcodesauge --barcode image.pngBarcode and QR detection via VNDetectBarcodesRequest. Returns the decoded payload plus the symbology (QR, EAN, Code128, PDF417, Aztec, and more).$ auge --barcode qr.png [QR] http://en.m.wikipedia.orgFacesauge --faces image.jpgFace detection via VNDetectFaceRectanglesRequest. Counts faces and returns normalized bounding boxes (x, y, width, height) you can draw over the image.$ auge --faces -o json group.jpg {"count":7,"faces":[ {"x":0.21,"y":0.38,"w":0.49,"h":0.38}, ... ]} 1Image in (any source)File path, stdin pipe (cat x.png | auge --ocr), --clipboard for NSPasteboard, or PDF (PDFKit handles text layer or rasterizes pages).2Vision request, on-deviceauge calls the matching VN…Request on your Mac's Apple Silicon (or Intel) GPU/Neural Engine. The model is shipped with macOS - no download, no API.3Structured outputPlain text, Markdown, JSON, NDJSON, or compact short-key JSON. Pipe it to jq, apfel, llm, or any UNIX tool.The data flowimage / pdf / clipboard / stdin→NetworkGuard (URLProtocol)→Vision framework (on-device)→JSON / Plain / Markdown / NDJSON → stdout NetworkGuard registers a URLProtocol that intercepts every http / https / ws / wss request inside the auge process and exits non-zero with a stderr message. If a future dependency, plugin, or accidental bug ever tries to reach the network, auge dies before the call lands.
Belt and suspenders. Input formats PNG · JPEG · HEIC · HEIF TIFF · BMP · GIF PDF (multi-page, text-layer or rasterize) NSPasteboard image (--clipboard) Stdin pipe (file paths or bytes) Analyses --ocr Text recognition (VNRecognizeTextRequest) --classify 1000+ labels (VNClassifyImageRequest) --barcode 12 symbologies (VNDetectBarcodesRequest) --faces Bounding boxes (VNDetectFaceRectanglesRequest) --all Every analysis in one pass Output formats plain Default; structured per mode json Pretty-printed structured output --compact Single-line JSON ndjson One JSON record per line md GitHub-flavored Markdown Barcode symbologies QR · Aztec · DataMatrix EAN-13 · EAN-8 · UPC-A · UPC-E Code 128 · Code 39 · Code 93 PDF417 · ITF (Interleaved 2 of 5) Plus URL/vCard/WiFi schema decoding OCR languages (BCP-47) Latin script: en, de, fr, es, it, pt, nl, sv, da, nb, fi, pl, cs, hu, tr, vi, cy, la CJK: zh-Hans, zh-Hant, ja, ko RTL: ar, he, fa, ur Other: ru, uk, sr-Cyrl, sr-Latn, el, hi, th, id Mix any of them with --langs a,b,c OCR controls --langs BCP-47 hints (priority order) --enhance Upscale tiny inputs before OCR --dpi PDF rasterization 72-600 (default 200) --prefer-embedded Use PDF text layer if present --clean FoundationModels post-pass (macOS 26+) Privacy guarantees 100% on-device · No API keys · No telemetry NetworkGuard URLProtocol blocks every http/https/ws/wss at runtime
Zero third-party Swift dependencies No model downloads (Vision ships with macOS) MIT license Platform & build macOS 10.15+ (full set on macOS 26 Tahoe) Apple Silicon or Intel Builds with Command Line Tools - no Xcode Swift 6.3 strict concurrency Hand-rolled CLI - no swift-argument-parser Every example below is processed by running the real auge binary (v1.1.0) on a public-domain document from Wikimedia Commons - at build time, on a Mac, with no network. Each card shows what auge produced: structured OCR text, classification labels with confidence, barcode payloads, face bounding boxes overlaid on the image, plus the full JSON output, on-device. Nothing is mocked or post-processed.61examples below14with face detection26with OCR text extraction10with QR / barcode decoding61with classification labels16distinct languages OCR'd face 1face 2face 3 faces 1948 Public Domain (Bell Labs PR photo, PD-US-no-notice)The transistor inventors - Bell Labs, 1948John Bardeen, William Shockley, Walter Brattain - the three Bell Labs scientists who invented the transistor in December 1947, photographed the next year. Every CPU on Earth descends from this workbench. auge detects all three faces.$ auge --all bell-labs-transistor.jpgA OCR no textNo text detected in this image.# Classify 10 labelspeople81%adult81%clothing71%necktie67%cord49%suit49%structure35%furniture35%table34%tableware26%▦ Barcodes noneNo barcodes or QR codes detected.☻ Faces 3 detected3 faces located. Bounding boxes overlaid on the image above.
face 1: x=0.273 y=0.700 w=0.123 h=0.155face 2: x=0.442 y=0.495 w=0.119 h=0.149face 3: x=0.609 y=0.651 w=0.129 h=0.162{} Raw JSON on-device output{ "file": "/Users/arthurficial/dev/auge-web/corpus/files/bell-labs-transistor.jpg", "metadata": { "on_device": true, "version": "1.1.0" }, "mode": "all", "results": { "barcodes": { "barcodes": [] }, "classify": { "classifications": [ { "confidence": 0.8076185584068298, "label": "people" }, { "confidence": 0.8076171875, "label": "adult" }, { "confidence": 0.7065683007240295, "label": "clothing" }, { "confidence": 0.67236328125, "label": "necktie" }, { "confidence": 0.4892578125, "label": "cord" }, { "confidence": 0.4873046875, "label": "suit" }, { "confidence": 0.3476963937282562, "label": "structure" }, { "confidence": 0.3450329601764679, "label": "furniture" }, { "confidence": 0.344970703125, "label": "table" }, { "confidence": 0.26153215765953064, "label": "tableware" } ] }, "faces": { "count": 3, "faces":
[ { "height": 0.1548747420310974, "width": 0.12341581284999847, "x": 0.27318617701530457, "y": 0.7003240585327148 }, { "height": 0.14936710894107819, "width": 0.11902691423892975, "x": 0.44163864850997925, "y": 0.4954269528388977 }, { "height": 0.16231800615787506, "width": 0.12934716045856476, "x": 0.6089426279067993, "y": 0.6511827111244202 } ] }, "ocr": { "lines": [], "text": "" } } }Wikimedia source classify 1968 Public Domain (NASA work)Earthrise - Apollo 8, 24 December 1968Bill Anders's photograph of Earth rising over the Moon, taken from Apollo 8 on Christmas Eve 1968. The most-reproduced photo of Earth ever taken. Galen Rowell called it 'the most influential environmental photograph ever taken.'$ auge --all earthrise.jpgA OCR no textNo text detected in this image.# Classify 10 labelsoutdoor58%celestial_body57%moon57%night_sky34%sky34%structure20%rocks20%liquid16%water16%frozen16%▦ Barcodes noneNo barcodes or QR codes detected.☻ Faces 0 detectedNo faces detected.{}
Raw JSON on-device output{ "file": "/Users/arthurficial/dev/auge-web/corpus/files/earthrise.jpg", "metadata": { "on_device": true, "version": "1.1.0" }, "mode": "all", "results": { "barcodes": { "barcodes": [] }, "classify": { "classifications": [ { "confidence": 0.5810793042182922, "label": "outdoor" }, { "confidence": 0.5659338235855103, "label": "celestial_body" }, { "confidence": 0.56591796875, "label": "moon" }, { "confidence": 0.3359375, "label": "night_sky" }, { "confidence": 0.3359375, "label": "sky" }, { "confidence": 0.1994629055261612, "label": "structure" }, { "confidence": 0.199462890625, "label": "rocks" }, { "confidence": 0.16483356058597565, "label": "liquid" }, { "confidence": 0.16483356058597565, "label": "water" }, { "confidence": 0.1648075133562088, "label": "frozen" } ] }, "faces": { "count": 0, "faces": []
}, "ocr": { "lines": [], "text": "" } } }Wikimedia source face 1face 2face 3face 4face 5face 6face 7face 8face 9face 10face 11 faces 1627 Public Domain (Hals died 1666)Banquet of the Officers of the St Hadrian Civic Guard - Frans HalsTwelve officers at a banquet, Haarlem 1627. Frans Hals founded the Dutch group portrait genre - eyes alive, faces individual. A perfect multi-face benchmark.$ auge --all hals-banquet.jpgA OCR no textNo text detected in this image.# Classify 10 labelsart84%painting84%people83%adult82%material55%textile55%crowd40%structure33%conveyance33%portal33%▦ Barcodes noneNo barcodes or QR codes detected.☻ Faces 11 detected11 faces located. Bounding boxes overlaid on the image above. face 1: x=0.102 y=0.641 w=0.054 h=0.078face 2: x=0.253 y=0.614 w=0.046 h=0.067face 3: x=0.426 y=0.429 w=0.049 h=0.072face 4: x=0.861 y=0.440 w=0.050 h=0.072face 5: x=0.592
y=0.624 w=0.039 h=0.057face 6: x=0.575 y=0.421 w=0.047 h=0.068face 7: x=0.739 y=0.422 w=0.051 h=0.074face 8: x=0.401 y=0.555 w=0.040 h=0.058face 9: x=0.858 y=0.636 w=0.049 h=0.071face 10: x=0.254 y=0.432 w=0.044 h=0.064face 11: x=0.689 y=0.694 w=0.044 h=0.064{} Raw JSON on-device output{ "file": "/Users/arthurficial/dev/auge-web/corpus/files/hals-banquet.jpg", "metadata": { "on_device": true, "version": "1.1.0" }, "mode": "all", "results": { "barcodes": { "barcodes": [] }, "classify": { "classifications": [ { "confidence": 0.8378981947898865, "label": "art" }, { "confidence": 0.837890625, "label": "painting" }, { "confidence": 0.82733553647995, "label": "people" }, { "confidence": 0.82080078125, "label": "adult" }, { "confidence": 0.5494153499603271, "label": "material" }, { "confidence": 0.54931640625, "label": "textile" }, { "confidence":