amebalabs/TRex

TRex is a macOS application that performs text recognition (OCR) on screen captures, clipboard content, and images, offering both local processing via Tesseract and cloud-based refinement using external LLM providers. It provides a comprehensive user interface for managing capture history, configuring automation shortcuts, and adjusting language or AI settings, alongside a command-line interface and Raycast integration for programmatic access. The system supports background watch modes, multi-region selection, and seamless updates via Sparkle, ensuring reliable text extraction and post-processing across various input methods.

9 changes.

This period focused on restructuring the application's architecture and user interface, introducing a comprehensive onboarding flow and a tabbed Settings window with dedicated sections for automation, OCR, and AI/LLM integration. Significant work included adding URL scheme support for remote control, implementing visual status indicators in the menu bar, and preparing the project for Mac App Store distribution with proper entitlements and localization structures.

8 changes.

This period focused on the development and release of TRex v2.0, introducing a modular architecture with TRexCore and TRexLLM packages to support LLM-powered OCR and post-processing. The work included building a comprehensive capture history system, a new command-line interface, and Raycast integration, while establishing automated build, signing, and Homebrew distribution pipelines.

A new Homebrew formula (cask) has been added for the TRex application, enabling users to install and manage the software via Homebrew. The cask is configured for version 2.0.0, includes automatic update support, installs the main application bundle, and provides a command-line binary named 'trex'. It also defines cleanup instructions for associated application support, cache, and preference files.

Formula · high confidence

Added four new shell scripts in the Resources/Raycast/TRex Commands directory that expose TRex's capture functionality to Raycast. These commands allow users to trigger text recognition (OCR) via the 'trex://capture' URL scheme, including specific variants for capturing from the clipboard ('trex://captureclipboard', 'trex://captureclipboardautomation') and running automations ('trex://captureautomation', 'trex://captureautomation').

Resources/Raycast · high confidence

A new Localizable.swift file has been introduced to organize localization resources. It defines a Localizable enum containing nested App and Preferences enums, providing a structured namespace for accessing localized strings within the application and preferences modules.

TRex/Resources/Localization · high confidence

Two new shell scripts have been added to the Scripts directory to support the build and update processes. The flatten-frameworks.sh script resolves App Store validation errors by removing improper Info.plist symlinks from TesseractCore and Leptonica frameworks. The generate-sparkle-keys.sh script automates the generation of EdDSA key pairs required for Sparkle code-signing updates, including instructions for configuring public and private keys.

Scripts · high confidence

Users can now view a history of text captures in a split-pane interface. The left panel lists entries with thumbnails, timestamps, and engine names, and includes a search bar to filter by text content. The right panel displays full details for a selected capture, including metadata like confidence and recognized languages, and provides a 'Copy to Clipboard' button to reuse the text. Users can also adjust the maximum number of stored history entries via a stepper and clear all history with a confirmation dialog.

TRex/App/UI/History · high confidence

The TRexLLM package now provides LLM-based text recognition and result refinement, allowing users to offload OCR and post-processing to external AI models. The system supports OpenAI, Anthropic, and custom OpenAI-compatible endpoints (such as local Ollama instances), with configuration for API keys, custom endpoints, and specific model selection. It includes an image preprocessor to optimize images for API transfer and a network checker to handle connectivity states. Users can enable LLM OCR and post-processing independently, with fallback to built-in OCR and configurable prompts for error correction and formatting.

Packages/TRexLLM · high confidence

A new command-line tool is added to TRex, allowing users to perform OCR captures and trigger automations directly from the terminal. The CLI supports three input modes: capturing the current screen, reading an image from the clipboard, or piping image data via standard input. When the automation flag is enabled, the extracted text is passed to configured Shortcuts. The tool requires macOS 10.15 or later and includes necessary entitlements for Apple Events automation and file access.

TRex CMD · high confidence

TRex Core now persists a local history of captures (including text, confidence, and thumbnails) via CaptureHistoryStore, and provides a FrozenScreenSelectionOverlay that freezes the display during region selection to prevent content from shifting. The package also adds LLM-based OCR and text post-processing engines (LLMOCREngine, LLMPostProcessor) with automatic fallback to built-in Vision or Tesseract engines, alongside a new InvocationMode enum to support multi-region capture and automation triggers.

Packages/TRexCore · high confidence

The Settings window has been restructured into a tabbed interface, introducing dedicated tabs for Automation, Custom Words, Tesseract, and (on non-MAS builds) AI/LLM. The Automation tab allows users to configure automatic URL opening and trigger macOS Shortcuts on captured text. The Custom Words tab provides a text editor to define specific vocabulary for improved recognition. The Tesseract tab enables users to toggle the Tesseract OCR engine and manage language downloads. The AI/LLM tab exposes settings for using external language models for both OCR and post-processing, including provider configuration and prompt editing. Existing General and Shortcuts tabs have been updated to support these new features, such as language selection logic that adapts based on the active OCR engine.

TRex/App/UI/Settings · high confidence

A change to existing behaviour in Resources — 9 commits, 6 files.

Resources · medium confidence · unverified

The application now includes a dedicated entitlements file for the Mac App Store build, enabling the required app sandbox, application groups, and Apple Events automation permissions. A new Info.plist defines the app's URL scheme, automatic update settings, and a Finder Service for text extraction, while a Credits.rtf file acknowledges third-party dependencies.

TRex/Resources · high confidence

The application now includes a dedicated utility to reliably open the Settings window across macOS versions 12 through 14, resolving previous issues where the settings option was missing or inaccessible. Additionally, a new file defines specific keyboard shortcut names for various capture actions (screen, clipboard, multi-region) and their automation triggers, establishing the configuration for these user-facing controls.

TRex/App · high confidence

The menu bar item now provides visual feedback during background operations: the icon pulses when the LLM is processing text and pulses at a different rate when watch mode is active, helping users distinguish between these two concurrent states.

TRex/App/UI · high confidence

The onboarding flow has been redesigned with a polished, branded UI featuring a three-step journey: a welcome screen highlighting key features, a quick setup page for configuring permissions (screen recording), language preferences, and app settings (shortcuts, launch at login, notifications), and a tutorial view demonstrating the capture process. The interface uses custom animations, brand colors, and modern component styles to guide users through initial configuration.

TRex/App/UI/Onboarding · high confidence

The Xcode project has been restructured to support a new modular architecture and updated user interface. A new 'TRex CLI' target has been added, enabling command-line usage of the tool. The main application now includes a comprehensive onboarding experience, featuring new views for setup, language selection, permissions, and tutorials, along with custom styling for text fields and toggles. Additionally, the project integrates the Tesseract OCR engine and the TRexCore package, while updating the KeyboardShortcuts framework to version 3.1 for compatibility with macOS 27.

TRex.xcodeproj · high confidence

TRex now supports a comprehensive set of URL scheme endpoints for remote control, including \capture\, \captureclipboard\, \captureautomation\, \capturemultiregion\, \showpreferences\, \showhistory\, \shortcut\, \startwatchmode\, and \stopwatchmode\. Additionally, the application has been updated to prevent the Settings window from opening automatically on launch, and it now correctly handles re-opening by showing and activating the settings if no visible windows are present.

TRex · high confidence

The TRexCore and TRexLLM Swift packages have been added, establishing the core library and LLM integration components. TRexCore now depends on TesseractSwift 1.2.1 and the local TRexLLM package, while TRexLLM depends on AnyLanguageModel 0.9.0. Both packages require macOS 14 and Swift 6.1, and include pinned versions for shared dependencies such as EventSource 1.3.0, JSONSchema 1.3.1, PartialJSONDecoder 1.0.0, swift-collections 1.3.0, and swift-syntax 602.0.0.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 56 → 55 (-1.1)
• Rubric changed (rubric-2026.09.11 → rubric-2026.09.18) — scores are not directly comparable.

• Code Health 94 → 94 (-0.9)
• Architecture 96 → 97 (+0.9)
• Maturity 49 → 49 (-0.0)
• Readiness 58 → 58 (+0.0)
• Security 46 → 46 (+0.0)

• Coverage not measured — Swift suite
• Documentation: contradicts the code (docs/release-notes/v2.0.1-beta.3.md)
• Duplicated block (10 lines × 2) (TRex/App/UI/Settings/LLMSettingsView.swift)
• Duplicated block (22 lines × 2) (TRex/App/UI/Onboarding/Views/TutorialView.swift)
• Duplicated block (25 lines × 2) (Packages/TRexCore/Sources/TRexCore/OCREngine.swift)
• Duplicated block (6 lines × 2) (Packages/TRexCore/Sources/TRexCore/TRexCore.swift)
• Duplicated block (8–9 lines × 2) (Packages/TRexCore/Sources/TRexCore/WatchModeOverlay.swift)
• Inconsistent naming for deletion operations. CaptureHistoryStore uses removeEntry, while LanguageManager and TesseractOCREngine use deleteLanguage. In Swift/iOS conventions, remove is often used for collection items (like array elements) and delete for persistent resources, but the inconsistency between removeEntry (generic) and deleteLanguage (specific) creates cognitive load.
• Inconsistent return types for equivalent data. MenuBarIcon exposes two methods to get the icon image, returning different types (Image vs NSImage). While Image might be a SwiftUI type and NSImage AppKit, exposing both as primary accessors is confusing. Usually, one should be the canonical source, and the other a conversion helper.
• MethodTooLong: GeneralSettingsView.body (TRex/App/UI/Settings/GeneralSettingsView.swift)
• Off the main sequence: TRexLLM
• Outdated: anylanguagemodel
• Redundant accessors. The type exposes a public property shortcuts which is an array, but also provides a method getShortcuts(). In Swift, accessing a property is the standard idiom; a get method is redundant and unidiomatic unless it performs significant computation or side effects, which is not indicated here.
• Repeated repair: Packages/TRexCore/Sources/TRexCore/ShortcutsManager.swift (Packages/TRexCore/Sources/TRexCore/ShortcutsManager.swift)
• Repeated repair: Packages/TRexCore/Sources/TRexCore/TesseractOCREngine.swift (Packages/TRexCore/Sources/TRexCore/TesseractOCREngine.swift)
• TooManyMethods: TRex (Packages/TRexCore/Sources/TRexCore/TRexCore.swift)

• 12 commits — 6 feature/other, 6 fixes

• Packages/TRexCore — 6 commits
• (root) — 3 commits
• docs/release-notes — 2 commits
• TRex.xcodeproj/project.pbxproj — 1 commit

• fix: fix(cli): exit after capture completes (#94)
• fix: fix(ocr): recover from Vision failures on macOS 27 (#95)
• fix: fix(shortcuts): update KeyboardShortcuts to 3.1 for macOS 27 (#102)
• fix: fix(tesseract): surface OCR failures instead of empty success (#98)
• fix: fix: recover empty OCR results with Tesseract (#96)
• fix: fix: restore clipboard after failed text writes (#97)
• change: chore(deps): bump TesseractSwift to 1.2.1 and drop stride workaround (#100)
• change: docs: add release notes for v2.0.1-beta.2
• change: docs: add release notes for v2.0.1-beta.3
• change: docs: refresh CLAUDE.md Tesseract and testing sections (#99)
• change: docs: update CHANGELOG.md, appcast and Homebrew formula for v2.0.1-beta.2
• change: docs: update CHANGELOG.md, appcast and Homebrew formula for v2.0.1-beta.3

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

amebalabs/TRex was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.