Changelog

What's new in Sygal

Every release, in one place. New features, improvements and fixes as they ship.

v1.5.7

Fix
  • OCR conversion using OpenAI (the default provider), NVIDIA, or Google Gemini could fail with an unexpected error — those providers had discontinued the underlying AI models; switched to their current replacements.
New
  • Checking for updates and downloading one now shows progress directly in the app (a spinner, then a progress bar with percentage), instead of only in the Windows taskbar.

v1.5.6

Fix
  • Fixed a packaging issue that could prevent the conversion engine from starting on some macOS versions, showing "conversion engine not responding" on every import.

v1.5.5

Improvement
  • Dropping files or folders to import now works anywhere in the window, not just on the library panel.
Fix
  • Exporting a document whose name contains characters like "/" or ":" (common in imported web page titles) no longer fails; very long names are now safely shortened too.
  • Importing a URL that returns no usable content (a site requiring login or JavaScript, like LinkedIn) now shows a clear error instead of an empty document marked "Done".
  • Deleting a document now removes its lines from the conversion log too, instead of leaving them behind until a manual clear or a restart.

v1.5.3

Improvement
  • The token savings panel moved from the bottom bar into a button next to the document tabs (Info / Markdown / JSON), with the gain still shown right on the button — no more visual mismatch with the conversion log below when only one of the two was open.
Fix
  • The conversion log now follows the app's display language instead of always showing in French.
  • The conversion log's header bar now reacts to a click anywhere on it to expand or collapse, not just the label text.
  • Panning and the scrollbar are now correct when zooming into an OCR'd image preview: dragging used to start the browser's native drag instead of moving the view, and the scroll couldn't reach the image's left/top edge.

v1.5.0

New
  • Zip archives can now be imported and exported either merged into one document or split into one document per file inside the archive — your choice, matching exactly what importing each file on its own would have produced.
  • Checking several documents and dragging one now moves the whole selection together, with a preview of the actual rows being moved instead of just the one under the cursor.
  • Double-clicking a document now checks it for batch actions, the same as Cmd/Ctrl-click.
  • RTF, MSG, EML, and TXT tables are rebuilt as real Markdown tables instead of a wall of pipe-separated text — including files found inside a zip archive.
  • XML now converts to structured, readable Markdown instead of a plain code block.
  • Every previewable format — HTML, XML, .xls, .epub, .ipynb, imported web pages — now zooms and pans like the rest, and Sygal remembers each document's zoom level and scroll position for the length of the session.
  • Cmd/Ctrl and +/- now zoom the document preview, from the keyboard or the menu.
  • A native zip browser lets you preview any file inside an archive without extracting it first.
  • Clicking a link or an image with a real URL in an imported web page or a YouTube transcript opens it in your browser.
  • A document's source path or URL is now clickable in the Infos panel, opening the file or the page directly — including for pages imported from the web, where it was missing entirely.
Improvement
  • The Markdown tab opens by default when a document is selected, instead of Infos.
  • YouTube transcripts are now captured in the video's own spoken language instead of defaulting to a fixed list of languages.
  • The Preview/Code toggle on Markdown documents reacts to a single click anywhere on it, the same as any other switch in Sygal.
Fix
  • Clicking "Show preview" while the window was too narrow to fit all three panels used to do nothing — it now widens the window automatically instead.
  • Reimporting a .md or .json file that was exported from a zip archive no longer fails.
  • Shift-click selection in the library now extends and shrinks correctly in both directions.
  • Various security hardening around external links and inter-process messaging.

v1.3.6

New
  • The NVIDIA NIM provider now uses a stronger vision model for OCR. Screenshots and documents that previously came back with a single line are now transcribed in full, accents included.
  • Hold Shift and click a second row in the library to select everything in between, the way Finder does.
Improvement
  • An image with no readable text now gets a proper description instead of a bare “no text found” line, and a widely recognizable character, brand or landmark is named when the identification is visually certain. Guessing anything about a real person remains off-limits.
  • The library only offers OCR where it can actually run: the actions are hidden when the active provider has no API key saved, since there would be nothing to click. Otherwise they stay exactly where you expect them, and the eye button remains the only way to hide them.
  • A document waiting on an API key reads the same whether it is an image or a scanned PDF, and stops looking like a failure once the key is saved. Retry is disabled, with an explanation, when it would only replay the same error.
  • Translate and the OCR actions are now clearly disabled, with an explanation, when the selected provider has no API key saved — instead of failing after the click.
  • When an OCR model stops after a couple of words, Sygal retries once and keeps the better result. If it stops again, the conversion log says so rather than silently storing a truncated transcription.
Fix
  • Translation no longer invents content. On a short document the model could write a whole story of its own, or translate the instructions instead of the text; it now receives the document as data it may only translate. If the result still looks incoherent, the original is kept untouched and the reason is shown.
  • A screenshot with no readable text no longer comes back as a single sentence with the description missing.
  • An .epub coming from Apple Books (stored as a folder rather than an archive) is imported as a single document again, instead of being split into one entry per internal file. Its images now show up in both the preview and the rendered Markdown.
  • Stray formatting produced by vision models is cleaned up: a code block wrapping the whole document, and tables returned as LaTeX, are converted back to plain Markdown.

v1.3.0

New
  • Added support for two new file formats: AVIF images and RTF documents (converted as plain text).
Fix
  • The token-cost comparison shown for a document (raw vs. converted) is now accurate no matter which conversion pipeline was used for a PDF.
  • Clear error messages for HEIC/HEIF photos and some unsupported AVIF files, instead of a generic error.
  • The image preview no longer crashes on those same formats.
  • RTF documents with tables now display as readable, properly separated text.
  • Email (.eml) previews are now visually consistent with the rest of the document preview.

v1.2.5

New
  • You can now deactivate your license directly from the app to free up a seat and activate it on another computer.
Improvement
  • New default NVIDIA OCR model (Nemotron Nano VL): significantly more reliable on images without text - portraits and illustrations now get a short, factual description instead of invented content, reliably in your language.
  • The new model is also faster and more consistent, with responses in 1-3 seconds once warmed up.
Fix
  • Import failures on Windows (e.g. drag & drop) are now clearly reported instead of failing silently, with a specific message per cause.
  • Screenshot files pasted into the library are now properly deleted from disk when removed from the library. Imported files are never affected.

v1.2.0

New
  • Images without readable text now return a structured, factual description instead of an empty result - for standalone images and for images embedded in documents alike.
  • Settings now shows a heads-up when the free default NVIDIA model is selected: it can be less reliable on images without text or with faces, and a stronger model can be set up as a custom provider.
Improvement
  • Much lighter app: about 450 MB of internal files that were never used at runtime have been removed from the installers.
  • Images sent to the NVIDIA OCR are now downscaled at constant quality instead of heavily compressed, preserving visual readability for the model.
Fix
  • Documents embedding the same image twice (common with InDesign PDF exports) are no longer transcribed twice: one OCR call, one block, at the right position.

v1.0.0

New
  • Sygal 1.0: convert any document, screenshot or link into clean Markdown and JSON, fully local.
  • New Plain text mode removes Markdown formatting for a lighter version to paste into an AI, while keeping the original and translated versions one click away.
  • Transcribe embedded images on demand: Word, Excel, PowerPoint and email files with untranscribed images get a one-click OCR button, and PDFs gain an 'Embedded images only' option in the OCR dialog, with API cost shown before you confirm.
Improvement
  • URL import is more robust: Brotli-compressed sites no longer return garbled text, imported pages keep their images and links (resolved to absolute URLs), and documents are reliably named after the page title.
  • Token counts are now faithful for every format: embedded images are counted as vision cost (Office, EPUB, emails, archives), PowerPoint tables and speaker notes are included, and the JSON count matches the exported file exactly.
  • Reorder your library by drag-and-drop all the way to the last position, including dropping in the empty area below the list.
  • Selected library row is easier to spot, in both light and dark mode.
  • Custom tooltips across the app, short and always fully visible, with keyboard shortcuts where available.
  • Cmd/Ctrl+F now searches the panel you are working in: the library when it has focus, otherwise the Markdown or JSON editor.
  • Documents awaiting conversion are highlighted in amber for every file type, clearly distinct from the selection color.
Fix
  • Removed a stray horizontal scrollbar in the document preview.
  • Files dropped directly onto an existing library row now import correctly, even when the list fills the entire panel.
  • OCR provider refusals (for example on images with faces) are now reported as a clear error instead of being saved as document content.
  • The library list no longer scrolls sideways; long file names are truncated with the full name shown on hover.

v0.9.97

New
  • Import from a URL (button or Cmd/Ctrl+L): web pages become clean Markdown, YouTube pages return title, channel and transcript when available, and RSS/Atom feeds become a readable list.
  • New file formats: Jupyter notebooks (.ipynb), ZIP archives (converted recursively) and legacy Excel (.xls).
Improvement
  • Cleaner in-app preview: spreadsheets, CSV and URL imports now render on a readable white page.
  • Design and interactive PDFs that used to preview blank now render correctly.

v0.9.96

New
  • A new Menu button in the sidebar lets you show or hide the Windows menu bar on demand (hidden by default).
  • Windows now has a dedicated File menu (Import, Save, Export, Quit), matching standard PC conventions.
Improvement
  • Windows updates now install silently and relaunch automatically, with no setup wizard - just like on macOS.

v0.9.95

Fix
  • Dropping a folder into the app now imports all of its compatible files instead of ignoring the folder.

v0.9.9

Improvement
  • The Windows menu bar now stays visible at all times, so features are easy to find without pressing Alt.
  • Settings, License, Check for Updates and About are now reachable directly from the Windows menu.
  • Windows menus are fully translated (French, Italian, German, Spanish), including About and Window.
Fix
  • macOS auto-update no longer fails with a connection error: updates now download and install correctly.

v0.9.7

New
  • "Convert via OCR" button on PDFs: rebuild tables, headings and reading order with a vision model when local extraction is not enough.
  • Page selector for OCR conversion: whole document, first page only, or a specific page.
  • Drag-and-drop reordering of the library — arrange your documents in any order, kept across sessions.
  • Toggle to show or hide the Convert via OCR button.
Improvement
  • Much better text extraction for PDFs with broken font encoding (invoices, forms): correct accents, no more garbled characters or barcode noise, proper euro sign, and top-to-bottom reading order.
  • OCR keeps the document's original language; labels the model adds now follow the app language instead of defaulting to English.
  • Redesigned auto-conversion control with an Auto ON/Off switch; the Convert all button is hidden when auto-conversion is on.
Fix
  • OCR "whole document" now converts every page instead of just the first one.
  • Fixed unwanted scrolling while selecting text in the editor with the search bar open.

v0.9.5

New
  • Licensing system with a 14-day free trial, no card required.
  • Trial banner with urgency colors: yellow, then orange, then red as the remaining days run out.
  • Native "License..." menu on both macOS and Windows.
  • Theme-adaptive logos that switch automatically between light and dark mode.
Improvement
  • Encrypted license storage using Electron safeStorage.
  • License validation on startup, with real-time revocation support.