• Updated: Privacy Policy (what crash reports you choose to share through Apple contain)

• Improved: At launch, ClariRec now also prepares on-device recognition for English-only text, so the first screenshot OCR of English text is ready sooner • Improved: For 30 minutes after an on-device recognition, ClariRec keeps text recognition prepared in the background, in case macOS clears its cached recognition models (this can happen often when disk space is low)

• Improved: Very large screenshots, such as a full 4K or 5K screen, are now recognized in sections — before, they could come back with little or no text • Improved: Cloud translations now appear as they are written instead of all at once at the end • Fixed: Clipboard OCR now tells you when the clipboard has no image instead of doing nothing • Fixed: Settings no longer lets two ClariRec actions share the same shortcut • Fixed: Quitting ClariRec while the on-device formula model is downloading keeps the download progress • Fixed: ClariRec now quits with ⌘Q even while a sheet, such as the model download, is open in Settings • Fixed: Settings opened from the menu bar now appears in front of other apps’ windows • Fixed: The “Launch at Login” switch now shows whether ClariRec actually opens at login • Updated: Privacy Policy (local records used to time the rating prompt)

• Fixed: The result window can now be moved — drag it by any empty part of the top or bottom bar • Improved: The result window reopens where you last dragged it instead of always in the center of the screen; the position resets when you quit ClariRec

• Improved: After a macOS update, ClariRec now prepares on-device text recognition in the background at launch, so your first screenshot OCR is ready sooner • Fixed: Starting a new screenshot while the previous recognition is still running now stops the old one instead of letting it slow the new one down • Improved: More reliable credential storage — if the Baidu OCR Secret Key can't be saved to Keychain, ClariRec tells you why and keeps your previous settings • Fixed: On macOS 27, icon menus in the result window, including “Copy or export formula”, have their proper accessibility names again

• New: Copy as Word Equation — paste into Microsoft Word and get an equation you can keep editing, not plain text • New: Export the corrected formula as a .tex or Markdown fragment, or as a transparent PNG at 2x or 3x for slides and documents • Improved: Copying and exporting now share one formula menu • Improved: When this month's 20 free on-device formula recognitions are used up, the upgrade screen now says so plainly • Updated: Privacy Policy and Terms of Use

• New: Correct a recognised formula without leaving ClariRec — edit the LaTeX source, watch the preview follow as you type, undo one step at a time, or reset to the recognised original • New: Syntax errors are explained in place, and commands the local preview cannot render say so instead of failing silently — either way the source stays editable and copyable • Improved: Raw, inline and display copy always take your corrected source; a format that would produce broken output is disabled with the reason shown • Fixed: Escape no longer closes the result window while the editor holds a correction you have not copied yet

• New: 20 free on-device formula recognitions every calendar month on eligible Macs — Pro removes the limit • New: Formula focus mode shows only the typeset result, with raw, inline, or display LaTeX copy • New: Five built-in formula examples you can open without downloading a model

• Improved: Updated the on-device inference engine behind on-device translation and on-device formula recognition

• New: Adjust the result window width from 360 pt to 560 pt — use the Display popover in the result window or the stepper in Settings • Improved: Faster result window updates

• Fixed: A completed purchase could briefly show as unpurchased while ClariRec re-checked your Pro status • Improved: A shorter first run — ClariRec now asks for screen recording access the first time you capture, instead of interrupting the intro with a trip to System Settings • Improved: More defensive window sizing on macOS 26

• New: Adjust the result text size from 12 pt to 20 pt with the Aa control in the result window, or with ⌘+ / ⌘- / ⌘0 • New: Formula recognition can now run entirely on your Mac. Download the engine once (~1.4GB), then turn screenshots of printed math into LaTeX with no API key — works offline. Requires Apple silicon with 16GB of memory or more • New: Choose your formula engine under Settings > Providers > Formula. The on-device engine is preferred once installed; a provider you set yourself is always respected • New: Adjust the result text size from 12 pt to 20 pt with the Aa control or ⌘+ / ⌘- / ⌘0 • New: Traditional Chinese interface • Improved: In the Chinese mainland App Store, on-device models now download from ModelScope • Improved: GLM-OCR recognized tables now come back as clean Markdown instead of raw HTML • Improved: Result window text stays sharp and is no longer clipped at larger sizes

• Fixed: Cloud provider responses, including recognized text and translations, were being cached to disk by the system. Caching is now disabled, and any existing cached data is removed on first launch • Fixed: The result window could crash the app on macOS 26 while resizing to fit its content • Fixed: Cancelling a screenshot selection could leave recognition stuck

• New: GLM-OCR cloud recognition — use your ZhipuAI BigModel or Z.ai API key for high-quality image text recognition; the app suggests a default platform based on your App Store region (BigModel for China, Z.ai for International) and you can switch between them at any time • Fixed: Saved API keys for translation-only providers, including SiliconFlow, GLM, and Mercury-2, now display correctly in settings • Fixed: Mercury-2 connection testing now validates the API key instead of relying on its public model-list endpoint

• Fixed: SiliconFlow translation with Qwen3-8B was extremely slow (~40 seconds for simple content) because the model's thinking mode was active by default; thinking mode is now disabled for translation requests, bringing response time in line with other providers • Improved: When Screen Recording permission has not been granted, pressing ⌥S no longer opens a redundant error window on top of the system permission dialog; the app now defers entirely to the macOS permission prompt and guides you to System Settings if needed • Fixed: Pressing ⌥T to translate clipboard text now reliably opens the Mini HUD translation flow again

• Fixed: DeepSeek is retiring its deepseek-chat model on July 24, 2026. ClariRec's built-in DeepSeek default is now deepseek-v4-flash, so DeepSeek translation keeps working without any action on your part • New: The DeepSeek translation model name can now be customized in Providers settings, so future upstream model changes no longer require an app update • New: SiliconFlow (SiliconCloud) translation provider

• New: Six more BYOK translation providers — Anthropic, DeepSeek, GLM, Grok, and OpenRouter — each ready with a built-in endpoint and default model; only an API key is required • New: The result window's recognized text is now directly editable — fix a typo and retranslate without rerunning OCR; an empty result is no longer a dead end, type or paste text to translate • New: Providers page redesigned around what you're configuring — text recognition, translation, or formula recognition — with inline setup and a clear status dot per provider • New: Translate selected text from other apps via macOS Services, or press ⌥T to translate clipboard text in a compact Mini HUD • New: Apple Intelligence translation on supported macOS 26 systems. No model download inside ClariRec and no API key required; availability depends on the system Apple Intelligence state • New: macOS 26 native document recognition for tables and lists, with Markdown table/list copy support • New: Switch translation target language directly in the result window and retranslate instantly • New: Technical text protection for LLM translation helps preserve URLs, inline code, file paths, snake_case, and lowerCamelCase identifiers • Improved: Clearer translation provider error messages for invalid keys, quota/rate limits, server errors, and unexpected responses • Fixed: System translation could hang indefinitely the first time a language pair needed an on-demand language pack; it now prompts to download the pack and retranslates automatically once ready • Note: V2.2 keeps the Pro boundary unchanged. New translation providers and macOS system capabilities are in the free layer; BYOK remote OCR execution and formula recognition remain Pro features

• New: On-device translation — download an optional local model (~1.1GB) to translate fully offline, with pause/resume support • New: Mercury-2 translation provider; macOS system translation is now a dedicated, explicit engine option • New: Translation and OCR engines are now independent. OCR reruns are one-off; translation engine switches persist as the new default • Changed: BYOK remote OCR execution is now part of ClariRec Pro; provider setup and BYOK translation remain free, and devices with prior remote OCR setup keep legacy access • New: Traditional Chinese added as a translation target language • Improved: Result window layout, and clearer indication when translated text may be truncated • Fixed: Local translation engine no longer risks a hang when the app quits during active use

• New: Formula recognition (Pro) — transcribe printed math formulas to LaTeX via BYOK VLM (Custom API, Qwen) • New: Local KaTeX rendering preview for formula results in the result window • New: Pro one-time purchase to unlock formula recognition

• New: DeepL translation provider with dedicated API support • New: OCR and translation engines can now be configured independently • New: Test Connection for BYOK providers before saving credentials • Improved: screenshot OCR targeting and capture geometry on multi-display setups • Improved: translation flow, target-language handling, and provider validation diagnostics

• New: Smart Text Cleanup repairs hyphenated line breaks and wrapped paragraphs • New: Detected regular tables can be copied as Markdown tables • Improved: text and Markdown exports now prefer table output when a regular grid is detected

Automation & translation refinements. • New: Shortcuts actions for clipboard OCR, file OCR, and text translation • New: `clarirec://` URL Scheme for clipboard OCR and text translation workflows • Improved: lower-noise translation experience for short text, same-language, and unavailable-system cases

Screenshot & layout refinements. • Fixed: settings sub-pages no longer get captured during screenshot OCR • Improved text layout processing for cleaner, more readable results

· Export Feature — Save recognized text as .txt or .md format · UI Refinement — Optimized menu spacing for crisp multi-language display · Architecture Upgrade — Better separation of concerns, improved stability, full test coverage

Version 1.3 focuses on translation and permission experience improvements. • Added DeepSeek translation preset for faster setup (pre-filled endpoint/model) • Fixed repeated Screen Recording permission requests during screenshot flow • Refined icon rendering for a cleaner, more consistent visual appearance