← Back to Blog

How to Search Text Inside Screenshots on Mac with OCR

Every day, software engineers, designers, and power users capture dozens of screenshots: a cryptic terminal compilation error, an AWS CloudWatch log snippet, a Figma component design, a Slack message thread, or a slide from a Zoom presentation.

The problem? Screenshots are dumb pixel blobs. Once captured, they get buried in your ~/Desktop or vanish into clipboard history. Days later, when you desperately need that specific error code or database connection string, standard search tools like macOS Spotlight cannot look inside your clipboard image history.

How Background Vision OCR Works Under the Hood

Rather than sending your sensitive screenshots to a remote cloud API, L2Cache utilizes native macOS Swift concurrency and Apple Silicon hardware acceleration:

1. Clipboard Capture Hook: Whenever a new image (PNG, TIFF, JPEG, HEIC, or WebP) is detected in the macOS pasteboard, L2Cache spawns a low-priority background Task on a dedicated Swift Actor.
2. Apple Vision Framework Execution: L2Cache invokes VNRecognizeTextRequest with recognitionLevel = .accurate, leveraging the Apple Neural Engine (NPU) for near-instant text recognition without impacting CPU or battery life.
3. SQLite FTS5 Full-Text Indexing: The extracted text, recognized language, and image dimensions are stored in an encrypted, on-device SQLite database powered by GRDB.
4. Instant Search & Copy: Press ⌥ + Space, type any word visible in the image, and L2Cache brings the screenshot to the top of your history with the recognized text ready to paste.

// Native Apple Vision Text Recognition Pattern
import Vision
import AppKit

func extractText(from image: NSImage) async -> String {
    guard let cgImage = image.cgImage(forProposedRect: nil, context: nil, hints: nil) else { return "" }
    
    let requestHandler = VNImageRequestHandler(cgImage: cgImage, options: [:])
    let request = VNRecognizeTextRequest()
    request.recognitionLevel = .accurate
    request.usesLanguageCorrection = true
    
    try? requestHandler.perform([request])
    let results = request.results as? [VNRecognizedTextObservation] ?? []
    return results.compactMap { $0.topCandidates(1).first?.string }.joined(separator: "\n")
}

4 Developer Use Cases for Searchable Screenshots

100% Local, Zero-Cloud Guarantee

Unlike browser-based converters or cloud clipboard tools, L2Cache operates with a strict zero-telemetry, zero-cloud architecture. Your screenshots remain on your Mac's NVMe drive, encrypted and protected behind macOS sandbox boundaries and optional Touch ID hardware locks.