OrbitOCR

Circle-to-Search for Windows — offline screen OCR and visual search in one hotkey.

OrbitOCR lives in your system tray. Press Ctrl + Shift + S (configurable), your desktop freezes across every monitor, and you can click or drag over any text to copy or search it, or draw a freehand circle around any image to search it with Google Lens. Text recognition runs entirely on the built-in Windows OCR engine — no cloud, no telemetry, no Tesseract, no Python.

Table of Contents

• Key Features
• Getting Started
  • Prerequisites
  • Install for end users
  • Build from source
  • Run the tests
  • Publish a standalone executable
• Usage Guide
• Configuration & Shortcuts
• Architecture & Tech Stack
• Contributing
• License
• Acknowledgments

Key Features

• 100% offline OCR — powered by the native Windows 10/11 Windows.Media.Ocr engine. Zero cloud APIs, zero telemetry, zero external runtimes. The explicit Search Google / Search with Lens actions are the only features that send data off-device, and only when you click them.
• Circle-to-Search interaction — inspired by Google Pixel.
  • Auto screen scan: on trigger, the whole virtual desktop is OCR'd in one background pass (typically ~150–250 ms) and every detected word becomes interactive.
  • Direct text interaction: hovering a word shows a soft glow and switches to an IBeam cursor; click a word or drag across a phrase to select it in reading order.
  • Region selection: drag a rectangle around any object, photo, or UI region to enter image mode, or hold Alt (or Shift) for a freehand lasso with a glowing cyan trail. Both show a live dimensions badge and draggable corner handles to refine the selection.
  • Dimming mask cut-out keeps your active selection at 100% brightness while the rest of the screen dims.
• Context-aware floating action pill
  • Text mode: Copy Text (with a synthesized chime and tray toast) and Search Google.
  • Image mode: Search with Lens (uploads the crop straight into Google Lens via the browser — no clipboard paste), Copy Image, Save Image (PNG/JPEG), plus Copy Text when text is detected inside the circle.
• Multi-monitor virtual desktop capture — captures all displays, including negative virtual-screen coordinates, and is PerMonitorV2 DPI-aware to prevent blur and coordinate drift.
• Global hotkey, no polling — registered through the Win32 RegisterHotKey API and a hidden message pump. Defaults to Ctrl + Shift + S; fully remappable from Settings (at least one modifier or an F1–F12 key is required to avoid accidental triggers).
• Background-first lifecycle — single-instance mutex, tray balloon on launch, and aggressive working-set trimming after every snip (idle footprint ~30 MB).
• OCR with small-text boost — crops up to 1500 × 1500 px are rescaled 2× with high-quality bicubic interpolation before recognition, significantly improving accuracy on small UI fonts.
• Language-aware — defaults to your Windows user-profile OCR languages, with a configurable picker over every installed recognizer language pack.
• Fluent dark UI — custom WPF design system (accent #60CDFF), DWM immersive dark title bar, and automation names/live regions for screen readers.

Getting Started

Prerequisites

Requirement	Details
OS	Windows 10 version 2004 (build 19041) or later, or Windows 11 — x64
Runtime (end users)	None. Use the self-contained build; no .NET installation required
Runtime (developers)	.NET 8.0 SDK or newer
OCR language pack	Installed via Settings → Time & Language → Language & region for any language you want to recognize (English ships with Windows)
Optional	Visual Studio 2022 17.8+ (workload .NET desktop development) or VS Code + C# Dev Kit

Install for end users

No installer and no admin rights required — OrbitOCR is a portable single executable (asInvoker manifest).

1. Download OrbitOCR.exe (Windows 10 2004+ / Windows 11, x64) from the Releases page.
2. Put it in any folder (e.g. %LOCALAPPDATA%\Programs\OrbitOCR) and run it. Since the binary is unsigned, Windows SmartScreen may show a warning on first launch — choose More info → Run anyway.
3. On first start the Settings window opens. Press Test or the global hotkey to try a snip.
4. Optional: enable Start with Windows in Settings to add OrbitOCR to your sign-in.
5. It now lives in the notification tray. Right-click the tray icon for Trigger snip, Settings…, About OrbitOCR, and Exit.

Or install with the Windows Package Manager:

winget install torpidno.OrbitOCR

The winget manifest is submitted to microsoft/winget-pkgs and resolves once that pull request is merged — until then, use the direct download above.

Build from source

git clone https://github.com/torpidno/OrbitOCR.git
cd OrbitOCR

# Restore, compile, run (opens the Settings window)
dotnet build
dotnet run

# Or start straight to the tray, skipping the Settings window
dotnet run -- --minimized

The app targets net8.0-windows10.0.19041.0 with WPF enabled, so build and run commands must be executed on Windows.

Run the tests

dotnet test tests/OrbitOCR.Tests.csproj

13 automated tests (MSTest) cover bitmap cropping and clamping, settings defaults and hotkey formatting, OCR language discovery, virtual-screen bounds, end-to-end OCR on clear and small dark-mode text, the Lens upload payload contract, PNG encoding, and XAML/resource smoke tests for every view.

The OCR end-to-end tests require a desktop session with at least one installed Windows OCR language pack.

Publish a standalone executable

Creates a self-contained, compressed, single-file OrbitOCR.exe with the .NET runtime bundled:

dotnet publish -c Release -r win-x64 --self-contained true `
  -p:PublishSingleFile=true `
  -p:IncludeNativeLibrariesForSelfExtract=true `
  -p:EnableCompressionInSingleFile=true `
  -o ./publish

Output: ./publish/OrbitOCR.exe — copy it to any folder, USB drive, or startup directory (publish/ is git-ignored).

Usage Guide

1. Trigger — press your global hotkey (default Ctrl + Shift + S), left-click the tray icon, or pick Trigger snip from the tray menu.
2. Wait for the scan — the desktop freezes, dims, and the top pill reports how many words were detected.
3. Interact:
   • Click any underlined word, or drag across several, to select text. A pill appears with Copy Text and Search Google.
   • Dragging anywhere that isn't text draws a rectangular selection. Hold Alt (or Shift) while dragging for a freehand lasso instead. Release to capture that region as an image; drag the corner handles to adjust it.
4. Act on the pill (copy, search, save), or press Esc to dismiss everything.

Everyday examples

Goal	Steps
Copy text from a video, PDF, or app that blocks selection	Hotkey → click/drag the words → Copy Text (or Ctrl + C)
Look up an error message	Hotkey → drag the message → Search Google
Identify a product, landmark, or plant	Hotkey → drag or circle it → Search with Lens
Save a region as an image	Hotkey → drag around the region → Save Image → choose PNG/JPEG
Reuse a screenshot in a chat	Hotkey → drag around the region → Copy Image → paste anywhere
Grab text without clicking the pill	Enable Auto-copy recognized text in Settings — the selection is copied the moment you release the mouse

Text mode (words detected under the cursor):

Action	Description
Click a word	Selects the single word
Drag across words	Selects the phrase in reading order, preserving spaces
Copy Text / Ctrl + C	Clipboard + chime + tray toast, then closes the overlay
Search Google	Opens https://www.google.com/search?q=… in the default browser
Esc	Cancels the snip immediately

Image mode (drag or circle a non-text region):

Action	Description
Search with Lens	Encodes the crop and hands it to Google Lens through your default browser via a self-submitting temp page (deleted after 2 minutes; stale files swept on startup). Falls back to clipboard + lens.google.com if the hand-off fails
Copy Image	Puts the cropped bitmap on the clipboard
Save Image	Save-as dialog; defaults to OrbitOCR_yyyyMMdd_HHmmss.png, PNG or JPEG
Copy Text	Appears only when text was detected inside the region; runs OCR on the crop with the 2× upscale boost
Corner handles	Resize the lasso bounding box before acting

Tray menu

Item	Description
Trigger snip	Same as the global hotkey (the configured shortcut is shown as its gesture)
Settings…	Shortcut recorder, behavior toggles, and OCR language
About OrbitOCR	Version and credits
Exit	Fully unregisters the hotkey and removes the tray icon

Left-clicking or double-clicking the tray icon also triggers a snip.

Configuration & Shortcuts

Keyboard shortcuts

Shortcut	Context	Action
Ctrl + Shift + S	Global (default, configurable)	Trigger a snip
Ctrl + C	Overlay, text selected	Copy the selected text and close the overlay
Alt (or Shift) + drag	Overlay, non-text area	Freehand lasso instead of the default rectangular selection
Esc	Overlay / shortcut recorder	Cancel the snip / stop recording
F1–F12	Global	Function keys are valid hotkeys without any modifier

Hotkey validation: a combination must include at least one of Ctrl, Shift, Alt, Win, or be a function key (F1–F12). Assignments that would swallow ordinary typing are rejected when saving. If Windows itself refuses a combination (another app already owns it), OrbitOCR keeps the previously working shortcut active and shows a warning instead of leaving you without one.

Settings window

Setting	Default	Description
Global shortcut	Ctrl + Shift + S	Click Record, press the combination, then Save. Test fires a snip immediately
Auto-copy recognized text	Off	Copies the selection to the clipboard as soon as the mouse is released over text
Sound feedback	On	Plays the Windows notification sound when text or an image is copied
Start with Windows	Off	Adds "OrbitOCR.exe" --minimized to HKCU\Software\Microsoft\Windows\CurrentVersion\Run, so sign-in boots straight to the tray
OCR language	Default (Windows user profile)	Any installed Windows.Media.Ocr recognizer language; falls back to the user profile, then to the first available language

Settings file

Settings persist as JSON at %APPDATA%\OrbitOCR\settings.json. The file is created on first save; a corrupt file silently falls back to defaults.

Key	Type	Default	Description
HotkeyCtrl	bool	true	Include Ctrl in the global hotkey
HotkeyShift	bool	true	Include Shift
HotkeyAlt	bool	false	Include Alt
HotkeyWin	bool	false	Include Win
HotkeyKey	string	"S"	Key name (Key enum name or a single character)
AutoCopyOnSnip	bool	false	Copy text automatically when a text selection ends
PlaySounds	bool	true	Play the copy chime
StartWithWindows	bool	false	Register/unregister the HKCU\...\Run entry (launches with --minimized)
PreferredOcrLanguage	string?	null	BCP-47 tag (e.g. "de-DE"); null follows Windows user-profile languages

Command-line flags

Flag	Effect
--minimized / /minimized	Start in the tray without opening the Settings window

Architecture & Tech Stack

OrbitOCR is a single-process WPF tray application. Services are composed by hand in App.OnStartup and communicate through events (HotkeyTriggered, TriggerSnipRequested, SettingsChanged, …), so the UI never talks to Win32 directly and the hotkey, tray, and settings subsystems stay testable and disposable.

Layer	Technology
Runtime	.NET 8 (net8.0-windows10.0.19041.0), C# 12, nullable reference types, implicit usings
UI	WPF (XAML + code-behind, no MVVM framework), custom Fluent dark design system in UI/Theme.xaml, DWM immersive dark title bar
OCR	Windows WinRT Windows.Media.Ocr.OcrEngine via Microsoft.Windows.SDK.NET projection
Capture	GDI+ CopyFromScreen across SM_*VIRTUALSCREEN metrics, System.Drawing.Common 8.0.8, high-quality bicubic 2× upscale for crops
Interop	Win32 RegisterHotKey/WM_HOTKEY, Shell_NotifyIcon, SetProcessWorkingSetSize, SetWindowPos, PerMonitorV2 DPI
Persistence	System.Text.Json → %APPDATA%\OrbitOCR\settings.json; HKCU\...\Run for startup
Lens hand-off	Generated temp HTML that multipart-POSTs the PNG to Google Lens' own upload endpoint, opened in the default browser (temp file lifetime 2 min, stale sweep on startup)
Tests	MSTest 3.1.1, Microsoft.NET.Test.Sdk 17.8.0, coverlet.collector 6.0.0

Project structure

OrbitOCR/
├── .github/workflows/build.yml     # CI: restore, build, and test on windows-latest
├── docs/                           # Landing page (GitHub Pages source) + README screenshot
│   ├── index.html                  # Static landing page (search + AI discoverable)
│   ├── llms.txt                    # Fact sheet for AI assistants and crawlers
│   ├── preview.png                 # README/landing screenshot
│   ├── sitemap.xml · robots.txt    # Search indexing
│   └── favicon.ico
├── LICENSE                         # MIT License
├── app.manifest                    # PerMonitorV2 DPI awareness, Win10/11 compatibility, asInvoker
├── OrbitOCR.csproj                 # net8.0-windows target, version 1.0.0, single-file publish spec
├── App.xaml / App.xaml.cs          # Composition root: tray lifecycle, single-instance mutex, memory trimming
├── Models/
│   ├── AppSettings.cs              # Hotkey, selection mode, sound/startup flags, OCR language
│   └── OcrExtractedResult.cs       # OCR result text + word bounding boxes
├── Services/
│   ├── HotkeyService.cs            # RegisterHotKey + hidden HwndSource message pump (MOD_NOREPEAT)
│   ├── ScreenCaptureService.cs     # Virtual-desktop capture, DIP-aware conversion, clamped cropping
│   ├── OcrService.cs               # OcrEngine wrapper: language selection, 2× upscale, word boxes
│   ├── LensSearchService.cs        # Self-submitting Lens payload page, temp lifecycle, cleanup
│   ├── TrayIconService.cs          # Shell_NotifyIcon, balloons, Fluent dark context menu
│   ├── SoundService.cs             # Windows notification sound on copy
│   └── SettingsService.cs          # settings.json load/save + startup registry
├── UI/
│   ├── Theme.xaml                  # Fluent dark design tokens, control templates, switches
│   ├── ActionMenu.xaml (.cs)       # Floating pill with contextual text/image actions
│   ├── OverlayWindow.xaml (.cs)    # TopMost frozen-desktop canvas, lasso, mask cutout, word layer
│   └── SettingsWindow.xaml (.cs)   # Hotkey recorder, toggles, OCR language selector
├── Assets/
│   ├── app.ico                     # Application (exe) icon
│   └── tray.ico                    # Tray icon, multi-resolution (16/32/48/64 px)
└── tests/
    ├── OrbitOCR.Tests.csproj       # MSTest project
    ├── UnitTest1.cs                # Unit + integration tests (cropping, settings, OCR, Lens payload)
    └── UiSmokeTests.cs             # Parses/lays out every view to catch XAML resource errors

Design notes

• No polling anywhere. The global hotkey arrives as a WM_HOTKEY message on a hidden window, and the tray icon as WM_TRAYICON callbacks.
• Capture → scan are decoupled. The full-screen OCR runs asynchronously after the overlay is shown, so the UI stays responsive while words stream in.
• Memory discipline. On overlay close, bitmaps and visual canvases are explicitly disposed/cleared, then a full GC plus SetProcessWorkingSetSize(-1, -1) trims the working set for an ultra-light tray footprint.
• Coordinates are kept in DIP space. Physical pixel word boxes are scaled to canvas coordinates, and crops are scaled back on capture — keeping selection accurate under DPI scaling and multi-monitor setups.

Contributing

Contributions are welcome — bug reports, fixes, and features alike.

Reporting bugs

Open an issue and include:

• Windows version/build and monitor layout (single or multi-monitor, DPI scale);
• steps to reproduce and what you expected;
• the OCR language selected and, if relevant, your %APPDATA%\OrbitOCR\settings.json;
• screenshots or a screen recording when the overlay misbehaves.

Check for existing issues first, and keep one issue per problem.

Pull requests

1. Fork the repository and create a topic branch (fix/lasso-resize, feat/tray-theme).
2. Make your change; keep diffs focused and match the existing style (file-scoped namespaces, nullable enabled, services single-purpose).
3. Build and run the full test suite — CI (.github/workflows/build.yml) runs the same restore/build/test flow on Windows for every push and pull request to main.
4. Open a PR describing what changed and why; link any related issue. This project follows a conventional commit style (feat:, fix:, docs:, test:).

Local development environment

git clone https://github.com/torpidno/OrbitOCR.git
cd OrbitOCR
dotnet build                 # compile
dotnet run                   # launch on the development desktop
dotnet test tests/OrbitOCR.Tests.csproj

• Windows 10 2004+ / Windows 11 with the .NET 8 SDK is required; WPF projects cannot be built on Linux/macOS.
• Visual Studio 2022 (workload .NET desktop development) or VS Code with the C# Dev Kit both work out of the box — no solution file is needed.
• The UI smoke tests and OCR integration tests need an interactive desktop session; run them locally, not in a headless agent.
• New dependencies should be justified — the app project deliberately ships with a single NuGet dependency (System.Drawing.Common).

License

OrbitOCR is released under the MIT License — free for personal and commercial use, with attribution. See LICENSE for the full text.

Acknowledgments

• Google Pixel's "Circle to Search" — the interaction model that inspired this project.
• Microsoft Windows OCR (Windows.Media.Ocr) — the on-device recognition engine that makes offline text extraction possible.
• Google Lens — the visual search destination for circled regions.
• Microsoft Fluent Design — the dark design language and accent palette used across the overlay and settings.
• MSTest, Microsoft.NET.Test.Sdk, and coverlet — the testing stack.
• Everyone who reports bugs, suggests features, and sends pull requests.