TRex

TRex v2.0.1 Beta 4 Release Notes

This beta improves screen selection on macOS 27 and prevents unsupported fast OCR from copying incorrect text.

Fixes

• Fixed frozen-screen selection causing other windows to animate or flicker on macOS 27. (#107)
• Restored the crosshair when entering or moving within frozen selection windows, including windows on other displays. Confirmation on a physical multi-monitor setup is still pending. (#112, #110)
• Prevented fast OCR fallback from returning Latin garbage for Arabic and other unsupported languages. When the accurate model fails, TRex now reports failure or uses a configured alternative instead of accepting incorrect fast-model output. (#111)
• Clarified the English language entry as American English. (#108)
• Fixed missing release notes in the Sparkle update window.

Improvements

• Updated supported-language discovery to Apple's modern Vision API on macOS 15 and later, while preserving existing language choices and macOS 14 compatibility. (#113)

Known Issues

• Apple's accurate OCR model can still stall intermittently. The Vision API update is not a fix for this issue. For Arabic, enable Tesseract in Settings and select/download Arabic; the reported sample was recognized correctly with that engine. (#106)
• The multi-monitor crosshair fix has automated regression coverage and single-display interaction testing, but still needs confirmation on a physical multi-monitor setup. (#110)

Requires macOS 14 or later. Includes support for Apple Silicon and Intel Macs.

TRex v2.0.1 Beta 3 Release Notes

This beta fixes recording keyboard shortcuts on macOS 26 and macOS 27.

Fixes

• Fixed the shortcut recorder in Settings → Shortcuts not capturing or saving shortcuts on macOS 26 and 27, where the control could also get stuck in the recording state. The KeyboardShortcuts library is updated to 3.1, which contains the upstream fix. Existing saved shortcuts are unaffected and load unchanged.

TRex v2.0.1 Beta 2 Release Notes

This beta fixes OCR on macOS 27, restores the command-line workflow, and makes Tesseract recognition work correctly for every language.

Fixes

• Fixed OCR failing or stalling on macOS 27: recognition now uses Apple's modern Vision API on macOS 15 and later, retries at the fast level when the accurate model fails, bounds every attempt with a timeout, and falls back to a configured Tesseract or AI engine instead of leaving the app stuck.
• Fixed Tesseract reporting success while leaving the clipboard empty. The underlying image-handling bug in TesseractSwift is fixed upstream in 1.2.1, and recognition failures are now reported as errors instead of silent empty results. This also makes Bengali, Greek, and other Tesseract-only languages recognize correctly.
• Fixed the trex command-line tool never exiting after a capture, which broke Alfred and scripting integrations. The CLI now exits with a proper status code, prints recognized text even when the clipboard write fails, and no longer races Shortcuts automation on exit.
• Fixed failed clipboard writes erasing the previous clipboard contents. The previous contents are restored when possible, without triggering macOS pasteboard privacy prompts on automatic captures.
• Fixed empty captures clearing the clipboard. Empty recognition results now still run table and document detection on the captured image before the capture is treated as empty.
• Fixed watch mode repeatedly re-running OCR and AI requests on unchanged frames whose recognition produced no text.
• Fixed OCR captures appearing unresponsive while recognition was running.
• Fixed empty automation and capture history inputs being processed.

Improvements

• Capture failures now show a notification instead of failing silently.
• The Tesseract language list distinguishes languages that are installed from those available to download.

Testing

• Added regression coverage for CLI exit behavior, Vision failure recovery and repeated recognition on macOS 27, Cyrillic support through the modern Vision API, clipboard restoration policy, watch-mode whitespace handling, padded-row and grayscale Tesseract recognition, empty-result recovery, and Bengali and Greek language exposure.

TRex v2.0.1 Beta 1 Release Notes

This beta focuses on update reliability, OCR correctness, and safer handling of captured content.

Improvements

• Production releases are now also published to the beta update feed, so users on the beta train continue receiving the newest eligible build.
• OCR engine routing, automatic language detection, custom words, and fallback behavior are more consistent across Vision, Tesseract, and LLM providers.
• LLM requests now use fresh sessions for each capture, validate custom endpoints, enforce image-size limits, and support Apple Foundation Models on macOS 26.
• Table output handles uneven rows, escaped Markdown, and embedded line breaks in CSV and TSV more reliably.
• App Store builds now normalize embedded frameworks and exclude unavailable third-party LLM controls and initialization.
• Onboarding, settings, capture history, menu behavior, accessibility labels, and drag-and-drop feedback have been refined for macOS.

Fixes

• Fixed Unicode text before a URL causing incorrect URL detection ranges.
• Fixed empty Vision results clearing the clipboard or capture history.
• Fixed OCR timeouts waiting indefinitely for operations that ignored cancellation.
• Fixed concurrent Tesseract requests racing through the shared engine.
• Fixed watch mode repeatedly processing blank captures and accepting unsafe output paths.
• Fixed temporary-file collisions and incomplete cleanup in the CLI, Shortcuts, and CLI installer.
• Fixed the CLI installer accepting invalid downloads and preserving stale executable permissions.
• Fixed release signing and notarization cleanup paths that could leave sensitive temporary files behind.
• Fixed screen-coordinate conversion on displays with different scale factors.

Testing

• Added regression coverage for OCR engine replacement, Unicode URL detection, watch-mode paths, table formatting, timeouts, LLM session isolation, endpoint validation, and image preprocessing.