OwlOCRDownload free
Release notes

OwlOCR changelog

A history of the customer-facing improvements that shaped OwlOCR. Recent releases include detailed notes; older entries collect the major milestones from the project archive.

v7.16.2
August 16, 2026

Safer, more reliable workflows

A maintenance update strengthens settings and OCR data handling while fixing nested Markdown placeholders in links.

  • Validated saved settings field by field while preserving existing defaults, migrations, and entitlement safeguards.
  • Checked native OCR responses, cache entries, and renderer-to-main messages before using their data.
  • Hardened file, URL, diagnostics, and export boundaries against malformed input without changing valid workflows.
  • Fixed inline code and image placeholders inside Markdown link labels so previews and plain-text cleanup no longer expose internal markers.
  • Expanded regression coverage for settings, OCR payloads, app communication, caching, Markdown rendering, and packaging dependencies.
v7.16.1
August 15, 2026

Complete searchable table text

Searchable PDFs now retain every recognized table word in its detected position while removing only unwanted table formatting.

  • Kept archival index entries searchable in their detected rows and cells while removing unwanted table formatting.
  • Flattened Markdown and HTML table formatting without deleting cell contents when formatted-table embedding is turned off.
  • Prevented exports from retaining only the first heading or block and avoided page-wide table selection boxes.
  • Kept right-margin references in their own positions and preserved each line’s detected width when replacing existing searchable text.
  • Hardened table cleanup for escaped and encoded pipes, inline code, entities, comments, quoted attributes, and multiline markup.
v7.16.0
August 13, 2026

Optional OvisOCR2 model

OvisOCR2 is now an optional local OCR AI model for structured pages, with verified downloads, isolated settings, and command-line support.

  • Added OvisOCR2 as an optional experimental OCR AI model that can process some structured pages faster with less memory.
  • Kept LightOnOCR 2 as the recommended default for degraded archival, multilingual, and scientific documents.
  • Downloaded Ovis model files separately from pinned revisions and verified their size and SHA-256 checksum before installation.
  • Applied model-specific prompts, runtime profiles, caches, and output cleanup when switching models.
  • Made the command line tool follow the OCR AI model selected in OwlOCR settings.
v7.15.0
August 12, 2026

Reliable table filtering and live OCR history

Plain-text archival tables now filter reliably, hidden PDF text follows each line more precisely, and live OCR keeps recent pages available for review.

  • Recognized archival index tables even when OCR AI returns ordinary text instead of Markdown or HTML markup.
  • Covered single-entry, short, wrapped, column-major, and merged-reference index layouts while preserving searchable headings and footers.
  • Removed excluded table text from existing searchable PDF layers instead of allowing stale embedded rows to survive export.
  • Matched hidden text to the complete detected line width and kept separate right-margin references in their physical positions.
  • Kept the viewed page plus 20 recently active pages in the live OCR text panel without retaining an unbounded document history.
v7.14.0
August 8, 2026

Accurately positioned searchable PDFs

OCR AI text now follows the physical words on each page, divider-less tables filter reliably, and wrapped-line behavior has one clear setting.

  • Mapped semantic OCR AI paragraphs onto their matching physical rows and columns without compressing them into one hidden line.
  • Kept whole words, headings, footers, and column order intact while rebuilding searchable PDF text layers.
  • Refreshed stale searchable text only when replacement geometry covers the complete page text.
  • Excluded divider-less Markdown tables, including blank cells, while preserving pipe-delimited math, shell commands, and code examples.
  • Consolidated the linked wrapped-line controls into a single setting under Fixes.
v7.13.3
August 7, 2026

Updated OCR runtimes and leaner downloads

Local OCR dependencies are refreshed, Mac packages exclude unused platform runtimes, and direct updates are validated before publication.

  • Updated LiteParse to 2.11.1 and the bundled llama.cpp inference runtime to b10310.
  • Refreshed RevenueCat, Electron packaging, networking, build, and website dependencies.
  • Excluded Linux and Windows LiteParse binaries from Mac packages to reduce download size.
  • Validated every update artifact, blockmap, size, and SHA-512 digest before publishing the update feed last.
  • Included the 7.13.2 searchable PDF text-placement, table-filtering, and compact scan-image fixes.
v7.13.2
August 5, 2026

Smaller, accurately positioned searchable PDFs

Searchable PDF exports now keep hidden text on its detected lines, reliably exclude OCR AI tables when requested, and preserve compact scan imagery.

  • Removed paragraph-wide PDF tags that could collapse several positioned OCR lines into one location.
  • Carried Markdown and HTML table roles through OCR AI layout normalization so the table-embedding setting works consistently.
  • Preserved literal table examples inside fenced, inline, and indented code as well as HTML comments.
  • Reused compatible compressed JPEG and JPEG 2000 scans, including color-profiled PDFs with small image masks.
  • Reduced the reported 21-page export from roughly 7.9 MB to 1.0 MB while retaining its original scan streams.
v7.13.1
August 4, 2026

More accurate searchable PDFs

Searchable PDF text stays aligned with the scanned page, and OCR AI table filtering now handles both Markdown and HTML output.

  • Kept every searchable OCR line anchored to its detected page position instead of stretching joined paragraphs across a large box.
  • Added standards-based paragraph semantics for compatible PDF readers while retaining accurate selection geometry.
  • Excluded HTML tables as well as Markdown tables when OCR AI table embedding is disabled.
  • Preserved literal table markup inside fenced code, inline code, and HTML comments.
  • Included wrap and text-cleanup settings in diagnostic logs for clearer export troubleshooting.
v7.13.0
August 3, 2026

Reliable long OCR AI runs

OCR AI now pauses safely between pages and keeps its local streaming connection reliable through long processing and extended pauses.

  • Changed Pause to finish active OCR AI pages before stopping the job at a safe page boundary.
  • Showed when a pause is waiting for active pages to finish, while keeping Cancel immediately available.
  • Resumed OCR cleanly from the page boundary without restarting completed work.
  • Prevented long quiet inference periods from timing out the local OCR AI streaming connection.
  • Kept extended pauses from breaking the connection to the local inference runtime.
v7.12.2
August 3, 2026

Accurate repeat OCR AI exports

Searchable PDFs now use fresh OCR AI text, naturally wrapped paragraphs copy cleanly, and the monochrome menu bar icon adapts on macOS Tahoe.

  • Replaced stale Apple Vision or earlier-session text with the fresh OCR AI result during PDF export.
  • Made the Markdown-table PDF setting take effect consistently on repeat exports.
  • Joined naturally wrapped prose for PDF copy, paste, and full-phrase search without changing the visible scanned page.
  • Preserved Markdown tables, headings, lists, code blocks, hard breaks, and recognized Unicode while joining prose.
  • Clearly identified repeat pages that reuse a cached OCR AI model result.
  • Preserved explicit text edits while refreshing unedited OCR output for export.
  • Made the black-and-white menu bar icon adapt to light and dark macOS Tahoe backgrounds.
v7.12.1
July 31, 2026

Cleaner, safer searchable PDF exports

OCR AI users can omit Markdown tables from the invisible PDF text layer, and very large exports now recover from an initial final-write failure.

  • Added an Advanced setting to omit OCR AI Markdown tables from searchable PDF text layers.
  • Kept visible PDF pages and OCR output unchanged when Markdown table embedding is disabled.
  • Retried large PDF finalization in bounded-memory chunks when the first streaming write cannot be validated.
  • Preserved prepared pages so an initial final-write failure does not discard hours of OCR work.
  • Added page-count, readability, and file-size diagnostics for PDF finalization failures.
  • Kept ordinary prose containing pipe characters in the searchable text layer.
v7.12.0
July 29, 2026

Faster, more reliable long PDF exports

Very large searchable PDFs now use less memory, retain live OCR output, and save reliably through the macOS sandbox.

  • Streamed scan-heavy PDF output with bounded memory to prevent failures during final writing.
  • Released the idle OCR AI runtime before memory-intensive PDF export while keeping active recognition work safe.
  • Kept live OCR text visible during PDF preparation without retaining every completed page in memory.
  • Upgraded recognition to Apple Vision Revision 3 and layout analysis to LiteParse 2.8.1.
  • Made newly published updates available without an artificial seven-day delay.
  • Preserved PDF metadata, page geometry, rotations, and searchable text through the streaming writer.
  • Restored Export Single Searchable PDF by staging output outside the save-panel destination folder and installing it atomically.
  • Kept existing files, aliases, and hard links protected when an atomic replacement is unavailable.
v7.11.6
July 24, 2026

Accurate OCR AI placement, compact PDF exports

OCR AI text now follows the source page’s lines and columns, while corrected scans preserve compressed imagery to avoid runaway file sizes.

  • Anchored OCR AI output to Apple Vision layout geometry instead of one page-sized text box.
  • Kept search selections and searchable-text placement aligned across rows, tables, and multi-column layouts.
  • Cached reusable page geometry together with OCR AI results for fast, correctly positioned repeat recognition.
  • Preserved original JPEG and JPEG 2000 scans where safe, adding sparse overlays only where page pixels changed.
  • Retained page rotations, vector details, and corrected searchable text while rebuilding scanned pages.
  • Restored previous text and layout when interrupted OCR did not complete.
  • Prevented cache clearing and late background work from overwriting newer results.
v7.11.5
July 18, 2026

Lower-memory exports for huge PDFs

Very large searchable PDFs now use far less memory, render at the correct resolution, replace stale OCR layers more reliably, and pause or cancel every export worker together.

  • Kept memory bounded while writing long searchable PDFs, including Separate PDF exports.
  • Rendered OCR pages at the requested pixel resolution to reduce unnecessary work without sacrificing accuracy.
  • Inspected embedded PDF text in small batches so very large documents remain responsive.
  • Replaced stale OCR text on scanned pages that also contain small vector or inline elements.
  • Preserved page geometry, metadata, and visible content while rebuilding searchable layers.
  • Paused, resumed, and cancelled every export worker together with reliable cleanup during immediate cancellation.
v7.11.4
July 18, 2026

Faster, cleaner searchable PDF exports

Searchable PDFs now write vastly faster while preserving mixed and complex pages, replacing stale OCR layers, and showing elapsed time and an end-to-end estimate with Pause/Resume controls.

  • Replaced repeated hidden-text fitting with one-pass layout, dramatically reducing PDF writing time on long and table-heavy documents.
  • Replaced stale searchable text on safe scanned pages so corrected exports no longer contain duplicate OCR layers.
  • Preserved mixed PDFs page by page, including clipped or masked scans, vector content, links, annotations, rotations, and page geometry.
  • Added exact elapsed time and an end-to-end estimate that includes PDF writing and finalization.
  • Added Pause and Resume controls for long OCR and searchable PDF jobs without counting paused time.
  • Kept diagnostics useful while recording timings and counts instead of recognized document text.
v7.11.3
July 18, 2026

Faster multipage OCR with live estimates

Multipage documents now recognize pages concurrently and show a rough time-remaining estimate, while OCR AI gains adaptive resolution, on-device caching, fast paths, simpler settings, and a newer local inference runtime.

  • Reduced OCR AI work with adaptive image resolution, a private on-device result cache, and safe embedded-text and blank-page fast paths.
  • Processed PDF and image pages concurrently while preserving document order.
  • Added rough time-remaining estimates for multipage OCR and PDF export.
  • Focused OCR AI settings on useful choices, including optional small-image upscaling and a reset to recommended defaults.
  • Allowed OCR AI to handle up to two pages at once while keeping memory use controlled.
  • Allowed the inference runtime to select supported Metal acceleration automatically.
  • Updated the bundled llama.cpp runtime to b10064 and made its reported build version accurate.
v7.11.2
July 15, 2026

Reliable Mac shutdowns

OwlOCR now closes promptly during Mac shutdowns and restarts instead of holding up the process or requiring a manual quit.

  • Allowed macOS shutdown and restart requests to close OwlOCR immediately.
  • Kept normal Quit behavior clean by stopping active OCR and local AI processes before exiting.
  • Applied the same reliable shutdown behavior across Intel and Apple silicon Macs.
v7.11.1
July 13, 2026

Less intrusive toolbar tooltips

Toolbar tooltips now get out of the way when a screen selection begins and can be disabled across the main toolbar.

  • Dismissed a toolbar tooltip as soon as its tool is activated so it cannot cover a screen selection.
  • Added a General > Toolbar setting to disable tooltips across the main toolbar.
  • Kept toolbar controls labeled for VoiceOver when visual tooltips are turned off.
v7.11.0
June 21, 2026

Automatic page orientation

OwlOCR can now detect and rotate sideways or upside-down pages before recognizing their text.

  • Added an opt-in four-way orientation check for turned pages.
  • Kept previews, OCR text boxes, and searchable PDF exports aligned with the detected rotation.
  • Supported automatic orientation with both Apple Vision and OCR AI while preserving manual page rotations.
v7.10.2
June 20, 2026

More reliable OCR results

Successful OCR results now return reliably even when macOS emits an unexpected native diagnostic message.

  • Reserved the native helper output channel for structured OCR responses.
  • Ignored and recorded unrelated native diagnostics without discarding a valid OCR result.
  • Made progress and final-response decoding safer across single-page, multipage, and PDF export workflows.
v7.10.1
June 7, 2026

Reliable CLI Pro access

The command-line tool now recognizes active Pro licenses reliably across direct-download and Mac App Store installations.

  • Recognized validated direct-download Pro licenses when running CLI commands.
  • Preferred settings for the matching distribution so stale settings from another installation no longer mask active Pro access.
  • Kept direct-download license tokens isolated from Mac App Store CLI entitlement checks.
v7.10.0
June 6, 2026

More reliable OCR AI

Local AI recognition now resists repeated text loops and uses an updated runtime, while scanner batches open in a more natural page order.

  • Matched the recommended LightOnOCR sampling and image settings across the app and command line tool.
  • Added a conservative repetition penalty to stop long or dense documents from falling into repeated generation loops.
  • Updated the bundled llama.cpp runtime from b9442 to b9542.
  • Kept unsuffixed scanner pages before numbered continuation files such as scan_001 and scan_002.
v7.9.0
June 3, 2026

Faster searchable PDF exports

Large searchable PDF exports now finish faster and provide visible progress while separate document files are written.

  • Reduced the cost of fitting hidden searchable text into narrow OCR boxes, which speeds up PDF writing for large batches.
  • Added page-level progress while writing searchable PDFs, including the current filename and document count.
  • Added cancellation support while searchable PDFs are being written.
  • Kept account numbers and similar text contiguous in the searchable PDF layer.
v7.8.1
June 2, 2026

Pro access and accessibility

Direct-download builds can recognize existing Mac App Store Pro access, while dialogs, controls, and navigation are clearer and more predictable.

  • Recognized existing OwlOCR Pro access from a Mac App Store installation when using the direct-download build on the same Mac.
  • Improved accessibility labels across settings controls, dialogs, links, and navigation.
  • Used native dialog behavior for onboarding and release notes to improve focus and dismissal.
  • Hardened dependency installation policy for safer release builds.
v7.7.2
June 2, 2026

Long PDF reliability

Longer PDFs now process more predictably in the app and Finder Quick Actions, with smaller searchable PDF exports and more careful page handling.

  • Fixed searchable PDF exports that could become much larger than the original document.
  • Preserved page geometry more carefully for rotated PDFs to avoid clipped or blank output pages.
  • Reduced temporary memory pressure while Finder Quick Actions process long documents.
  • Kept multipage PDF OCR sequential within each document for the same stable behavior as single-page recognition.
v7.7.1
May 31, 2026

OCR AI performance controls

Local OCR AI can now be tuned for the balance of memory use, speed, and recognition quality that fits your Mac.

  • Added smaller Q3_K_S and Q2_K model variants alongside the balanced Q4_K_M default.
  • Added an Advanced options section for llama-server context size, GPU layers, CPU threads, batch sizing, and Flash Attention.
  • Added memory mapping and RAM locking controls for local runtime tuning.
  • Kept the app and command line OCR paths aligned with the selected model variant.
v7.7.0
May 31, 2026

Cleaner OCR, better control

Long-running OCR work is easier to follow and easier to stop, with a new set of text cleanup tools for the result.

  • PDF text extraction preserves more accurate document text when it is available.
  • OCR AI streams completed pages into your document while it works.
  • Stopping OCR AI is faster and keeps the local runtime ready for your next request.
  • OCR All resumes unfinished pages instead of repeating completed work.
  • A new Fixes section can join wrapped lines, repair split words, normalize punctuation spacing, and collapse repeated blank lines.
v7.6.0
May 26, 2026

Purchase and CLI reliability

This release makes Pro access easier to understand and command line workflows more dependable.

  • Mac App Store subscriptions are checked against current entitlements more accurately.
  • The redesigned Pro panel makes plan selection, checkout, and restoring purchases clearer.
  • The command line tool handles relative image and PDF paths more reliably.
  • Permission and file-format errors from the command line tool are easier to diagnose.
v7.5.0
May 23, 2026

OCR AI workflow fixes

OCR AI gained a smoother command line path and more predictable ordering for document batches.

  • Use `--ocr-ai --markdown` to write Markdown output from Terminal.
  • Numbered scans are loaded in natural filename order.
  • Finder Quick Actions and packaged Mac App Store builds received reliability fixes.
v7.4.0
May 13, 2026

Electron rewrite support fixes

A maintenance release tightened the new app after the OwlOCR 7 rewrite and refreshed its dependencies.

  • Fixed support regressions introduced during the Electron migration.
  • Updated packaged dependencies for the modern app.
v7.3.0
May 11, 2026

Local OCR AI

OwlOCR 7 added an optional second OCR engine for difficult pages while keeping document processing on your Mac.

  • OCR AI can handle messy screenshots, low-quality scans, dense documents, and unusual layouts.
  • Apple Vision remains the fast default for text boxes, highlights, and searchable PDF export.
  • Model and runtime downloads are managed from Settings.
  • The local OCR AI runtime is warmed up ahead of use for a faster first request.
v7.2.0
May 8, 2026

Pro purchase restoration and CLI

The modern app gained a supported automation path and more dependable Mac App Store purchase restoration.

  • Added the OwlOCR command line tool for OCR automation.
  • Moved Mac App Store purchases to StoreKit 2 entitlement checks.
  • Fixed the OwlOCR Pro purchase restore flow.
v7.1.0
May 7, 2026

A clearer upgrade path

The classic OwlOCR app became free to use while the new app added a clearer path for restoring and managing Pro access.

  • Added a restore prompt for existing customers.
  • Restored and refined the Pro settings pane.
  • Updated onboarding copy for customers moving from the classic app.
v7.0.0
April 30, 2026

A new foundation

OwlOCR was rebuilt as an Electron app with focused native Swift helpers for the Mac-specific work.

  • Rebuilt the main workspace, settings, menu bar workflow, and document state in the new app.
  • Kept native Apple Vision OCR, PDF rendering and export, screen capture, and Continuity import.
  • Preserved the three Finder Quick Actions: Extract Text, Create Searchable PDF, and Create Text File.
From the archive

Earlier milestones

The project started in 2020. These highlights are distilled from the historical commit record, with smaller fixes folded into the releases they improved.

2023 - early 2026

Refining the classic app

  • Added automatic OCR when opening a document and an option to open PDFs after screen capture.
  • Added OCR algorithm revision selection as Apple expanded Vision language support.
  • Added more line spacing and minimum text size controls.
  • Modernized launch-at-login behavior and expanded the available OCR languages.
2022

OwlOCR 5 and a major performance pass

  • Shipped OwlOCR 5 with fixes for file access and QR code scanning.
  • Refactored concurrency, loading, and document models to improve responsiveness.
  • Reduced PDF memory use and added parallel file loading.
  • Stabilized the Finder extensions after the broader refactor.
2021

Finder workflows and text cleanup

  • Added custom OCR words, post-processing rules, and smart quote handling.
  • Introduced Finder Quick Actions for clipboard text, text files, and searchable PDFs.
  • Added onboarding, page controls, separate PDF export, and adjustable text size.
  • Shared settings between the main app and Finder extensions.
2020

From Textify to OwlOCR

  • Started as Textify in January and became OwlOCR in February.
  • Added rotated PDF support, iPhone import, configurable screenshot shortcuts, and multi-document workflows.
  • Made OCR results editable, added language selection, and accepted files dropped onto the Dock icon.
  • Introduced Quick Export for one-step screen capture to searchable PDF.