kesit

Native macOS agentic code-search — extract, clean, and translate code from visual sources using on-device AI.

Website • Privacy • Support

---

Kesit captures your screen, extracts code using Apple Vision OCR, cleans it with a local LLM, and lets you transform it with agentic "Skills." No data ever leaves your Mac.

How it works

Screen Capture  →  OCR (Vision)  →  LLM Cleanup  →  Agent Skills

1. Capture — select a code region via a global hotkey (⌥ Space) and floating Lasso overlay
2. Extract — recognise code text using Apple Vision (VNRecognizeTextRequest)
3. Refine — diff-resolution preprocessing, then an on-device LLM "Janitor" cleans OCR artifacts, fixes indentation, ensures syntax validity
4. Transform — apply agentic Skills: "Translate to Swift 6", "Add documentation", "Format as Markdown"

Features

Capture & Input

• ✂️ Global Lasso — ⌥ Space triggers a floating selection overlay anywhere on screen
• 📐 Multi-Region Capture — select multiple code regions and stitch them into one extraction
• 📋 Clipboard Monitor — auto-detects images in clipboard and offers instant extraction
• 🌐 Source Window Context — automatically captures app name, window title, and URL from browsers

Intelligence & Processing

• 🔒 100% On-Device — no cloud calls, no telemetry on your screen content
• 🧠 Apple Intelligence — leverages FoundationModels for code cleanup and transformation (graceful fallback on older systems)
• 🔎 Language Detection — auto-detects 17 programming languages (Swift, Python, JavaScript, TypeScript, Java, Kotlin, Rust, Go, C#, C++, C, Ruby, PHP, HTML, CSS, SQL, Shell) with confidence scores
• 🔨 Compilation Sandbox — Swift syntax is validated locally via SwiftSyntax parser (no Xcode dependency required)
• 🔀 OCR Diff View — side-by-side comparison showing exactly what the LLM Janitor fixed
• 🤖 Agentic Skills — prompt-based transformations: translate languages, add documentation, reformat — all running locally
• ✏️ User Feedback Loop — edit incorrect OCR results; edits are stored for few-shot learning context

Library & Organisation

• 🔍 Searchable Snippet Library — full-text search, language filters, and tags across all captures
• 📚 Snap History — sidebar groups captures by source window title (e.g., "YouTube — Swift Tutorial")

Export

• 📦 Export Anywhere — Clipboard, Markdown file, or Swift Playground
• 🖥️ Menu Bar Quick Access — persistent status item for one-click copy and capture

Design & Architecture

• 🎨 Pitch Black UI — terminal-modernist design with San Francisco Mono typography
• ⚡ Built with Swift 6 — strict concurrency, SPM modular architecture, async/await throughout
• 🧩 Modular SPM Packages — KesitCore, FeatureLasso, FeatureIntelligence, FeatureStorage, FeatureUI, SwiftEnsemble
• 🔗 SwiftEnsemble Pipeline — chainable node architecture for orchestrating on-device ML pipelines

Requirements

Minimum	macOS 15.0+ (Sequoia)
Apple Intelligence	macOS 26+ (optional — deterministic fallback on older systems)

Download

🍎 Mac App Store

---

Built with SwiftEnsemble — a Swift-native framework for orchestrating on-device ML pipelines.