3.2026.725

New Features

• New CAD drawing viewer: open DWG and DXF files directly, inspect vector entities, layers, and layouts, and use zoom, pan, entity selection, and distance measurement. When DWG vector parsing fails, OpenFiles also attempts to show the embedded preview.
• Agnes AI provider: built-in presets for Agnes text, image, and video models, including vision, tool use, thinking mode, reference-image generation, and multiple video durations, resolutions, and aspect ratios.

Improvements

• Unified file-format catalog across desktop and web: 350+ formats now share the same searchable, categorized detail experience. Desktop settings can also open files directly from a format or its associated app.
• Refined desktop interactions: polished the workspace, Hub, settings, AI conversations, and data-editing dialogs with denser narrow-window layouts, clearer selection states, and larger interaction targets.
• Refreshed website and administration UI: improved the homepage, navigation, downloads, authentication, feedback, Dashboard, and admin pages with more responsive layouts, stronger anchor navigation, and clearer conversion actions.

Fixes

• Fixed spreadsheet edits that could corrupt adjacent cells when double-clicking or clearing a value, and restored correct editing and rendering for booleans, formulas, and charts backed by formula results.
• Fixed only one image viewer rendering in split mode, RAW decoding workers failing in packaged builds, and rich Markdown heading text missing from the document outline.
• Fixed macOS Keychain prompts when saving custom AI providers and desktop feedback submissions being incorrectly discarded as too fast.
• Fixed encrypted-archive password detection, incomplete files left behind after failed extraction, and metadata handling for some archives. DJI SRT telemetry timelines and flight durations are also more accurate.
• Fixed incorrect desktop usage attribution and retry storms after failed tracking requests, making activity statistics, sign-out, and cross-device sessions more reliable.

changelog

3.2026.718

New Features

• Spreadsheet viewer upgraded to an editor: Excel workbooks can now be edited and saved directly. Formulas are recalculated as cells change, while saving preserves the original workbook structure wherever possible.
• New changelog browser: the desktop settings and website changelog now include version search, a release list, and pagination, making specific releases easier to find in a long history.

Improvements

• Excel previews closer to the original: improved rendering for cell styles, borders, merged cells, row and column sizes, frozen panes, hyperlinks, comments, data validation, conditional formatting, sparklines, images, shapes, and charts. Hidden worksheets are no longer shown.
• More consistent editor and chart themes: viewers including Markdown, Mermaid, PlantUML, Graphviz, Jupyter, palettes, flight logs, and SRT telemetry now follow the active light or dark theme more accurately. Exported Markdown HTML also uses the current theme.
• Clearer workspace navigation: the bottom gallery shows full tooltips for truncated filenames and keeps native horizontal scrollbars from covering names. File tree, view-mode controls, and narrow-window layouts are also more stable.
• More reliable Markdown editing: switching between source and WYSIWYG modes now synchronizes the latest content first, preventing rapid switches from leaving content or the table of contents behind.
• Lower startup memory usage on Windows: workspace restore now initializes only visible tabs. Inactive tabs load when first opened, preventing multiple viewers from allocating memory at startup.

Fixes

• Fixed chat database upgrades blocking application startup. Failed upgrades now retry in the background while preserving a recoverable startup path.
• Fixed search indexing for large folders occupying the main process. Root indexing now runs in a dedicated worker to reduce startup and file-browsing stalls.
• Fixed long labels overflowing in Mermaid, PlantUML, and other SVG diagrams, with stronger SVG sanitization, wrapping, and theme-color handling.
• Fixed PowerShell file-association generation, OpenNext static-asset packaging, OFD and Markdown viewer lifecycle issues, and several dark-theme rendering details.
• Fixed custom Provider model settings not being saved reliably. Newly entered models now receive a valid name and default selection automatically, and editing model IDs no longer drops input focus.

changelog

3.2026.705

New Features

• PlantUML editor and Markdown rendering: OpenFiles now opens `.puml` / `.plantuml` files with a dedicated PlantUML editor, bundled local rendering, preview controls, and file icons. Markdown code blocks can render PlantUML diagrams alongside Mermaid and Graphviz previews.
• Viewer search across more formats: document-like viewers now expose a consistent toolbar search entry, including PDF, PPTX, DOCX, EPUB, HTML, OFD, sheets, Graphviz, code, Mermaid, and PlantUML paths.
• Homepage and shell polish: the website homepage and desktop shell motion were refreshed with lighter transitions and more stable styling.

Improvements

• Better PPTX fidelity: PowerPoint rendering now handles more bullet/list variants and avoids text-frame clipping in cases reported by users.
• More robust diagram previews: Mermaid lightbox behavior was tightened, PlantUML rendering works across packaged/dev builds, dark SVG output is normalized, and previews fit their viewport more reliably.
• Markdown editor round-trip fixes: template placeholders such as `${...}` inside links are preserved instead of being mistaken for inline math.
• Admin analytics day windows: admin stats and analytics now use Shanghai natural-day semantics for "today" while keeping longer range windows rolling.

Regression Testing

• Verified the unsigned macOS arm64 build with Computer Use against the 3.2026.630-to-3.2026.705 changes: PlantUML editing/rendering, PPTX rendering and search, PDF search, Markdown/Mermaid rendering and search, and packaged app startup from the built `release/mac-arm64` app.
• Added and refreshed focused tests for PlantUML support, viewer search primitives, PPTX parsing/rendering, Markdown diagram behavior, homepage polish, and admin analytics semantics.

Fixes

• Fixed OpenNext/Next build compatibility around the PlantUML engine asset and backend deployment.
• Fixed PlantUML empty states, development engine loading, dark-theme preview color handling, and fit/refresh behavior.
• Fixed several shared UI overlays so dialogs, popovers, selects, sheets, and tooltips behave consistently after the viewer-search and shell updates.
• Fixed Windows packaging with npm 11 by vendoring the IFC loader assets needed by the model viewer and adding a manual node_modules dependency traversal fallback for electron-builder.
• Fixed CodeMirror-based viewer search so the search input is focused before the browser/editor find panel opens across code, Graphviz, Markdown, Mermaid, and PlantUML editors.
• Fixed ebook navigation crashes caused by invalid CFI indexes during relocation, and kept the table of contents plus previous/next buttons responsive after relocate errors.
• Fixed ebook state reset when switching files from the bottom gallery so the newly selected book can refresh, navigate, and open its table of contents correctly.
• Fixed ebook toolbar progress labels so reflowable books show percentages instead of Foliate Loc counters, while fixed-layout books such as CBZ keep page-like counts.

changelog

3.2026.630

New Features

• Paged Path Explorer loading: large folders now load in pages, keeping browsing responsive while directories with many files continue to stream in.
• PDF and ebook bookmarks: PDF and EPUB readers now include bookmark support, making it easier to mark important pages and return to reading context later.
• DJI mobile DAT previews: DJI mobile flight-record DAT files can now be parsed and previewed, expanding drone log coverage beyond the previously supported DAT variants.

Improvements

• Safer app startup and recovery: OpenFiles now has a global renderer error fallback and stronger main-process error handling, so unexpected UI failures degrade more gracefully.
• More reliable packaged media support: packaged ffmpeg/ffprobe lookup is hardened and verified during release builds, improving media playback and conversion reliability in signed builds.
• Homepage and local development polish: the website hero animation is lighter, local development path handling is smoother, and dependency prebundling now includes the Markdown serializer path used by the editor.
• Dependency refresh: core dependencies and lock metadata were refreshed, including the Tailwind WASI lock data used by the desktop build.

Regression Testing Fixes

• Unsigned macOS build sweep: completed a 3.x regression pass covering paged Path Explorer loading, Hub search, multi-window/split view, PDF/EPUB, Markdown, media, archives, CSV, and SQLite paths.
• Packaged path and startup dispatch fixes: fixed startup file dispatch waiting for the app registry, packaged renderer asset resolution, packaged Lottie WASM loading, relative import casing, and ASAR archive detection when opened through Path Explorer in macOS/dev flows.
• Viewer stability hardening: fixed SQLite grid crashes and stale cache release, gallery ResizeObserver loops, global fallback tab closing, PDF precise-summation polyfill, OFD preset font requests, and 3D Tiles loading.

Fixes

• Fixed macOS file-open flows so files opened from Finder or the system file manager bring the existing OpenFiles window forward, including minimized-window and cold-start cases.
• Fixed RAR "extract all" fallback behavior, OFD XML parsing through the declared parser, and release build import casing issues.
• Fixed Markdown editor regressions around image lightbox event scope and footnote definition layout.
• Fixed OpenFiles built-in provider storage so it avoids unnecessary keychain prompts.
• Fixed path explorer edge cases, file tree drag payload typing, and several local-service error surfaces.

changelog

3.2026.626

New Features

• Supported formats settings: Settings now includes a supported formats page for browsing file types, categories, and default opening capabilities in one place. Richer format-specific icons also make files easier to scan and identify.
• Mind map editing upgrades: the mind map app has grown from a viewer into a fuller editor, with new-file templates, layout preferences, node editing, notes, summaries, history, keyboard navigation, plus import, save, and export support for XMind, MindManager, MindNode, FreeMind, SimpleMind, Markdown, and related formats.
• Richer workspace inspector: the file info drawer now shows richer image, RAW, and Exif/EXIF metadata, including image dimensions and more useful previews. The workspace gallery shortcut also makes browsing media-heavy folders faster.

Improvements

• Faster image and RAW previews: optimized image loading paths and RAW startup, added thumbnail/full-decode caching and cancellation, removed the exifr dependency, and reused exiv2 metadata reads so large images, RAW files, and rapid file switching feel lighter.
• Markdown diagrams and exports: Mermaid, Graphviz, and exported HTML now follow the active theme with more coherent colors. Lightbox contrast and preview spacing were polished for steadier readability in dark mode.
• More reliable workspace opening flows: improved cold-start file association opening, the Explorer new-file dialog, replaced-path tab syncing, Markdown refresh/reload behavior, the Hub close button, workspace restore state, and packaged startup logo rendering.
• Release and engineering quality: GitHub Release uploads now route through Actions, while a large set of source, test, and script paths moved to snake-case with fixture paths and build references updated for better long-term path compatibility.

Fixes

• Fixed stale window state being restored after close, incomplete state persistence when the final workspace closes, deferred restored tabs being dropped, and edge cases in collapsed rails and gallery layout.
• Fixed stale image decode jobs writing late results, RAW preview startup stalls, repeated metadata reads, and missing image dimensions in the info drawer.
• Fixed a set of mind map editing details around image nodes, folded node IDs, default layout, topic details, pan selection, and image ordering.

changelog

3.2026.621

New Features

• Multi-window workspaces: OpenFiles can now open new independent windows from the current workspace, while syncing theme color, language, window sessions, and restore state. New windows are staggered to reduce overlap, and the title bar and sidebar now provide more direct workspace shortcuts.
• Hub Window: the standalone Hub window has landed, with polished launch / close behavior, default sizing, shadow treatment, file-row icons, collapsed-sidebar access, and workspace continuity. It now works well as a control center for quickly opening files, apps, and locations.
• Quick Look Window: added a standalone Quick Look preview window. Viewers render in preview mode, previews can continue into a workspace window, and the window title, error states, and multi-display placement are improved.
• Split View: workspaces now support resizable split groups. App reuse, close-and-merge behavior, boundary contrast, layout restore, and migration are more reliable, so multiple files can sit side by side in one window.
• Path Explorer upgrades: added folder favorites, in-archive browsing and extraction, selection of newly created items, arrow-key navigation, continued navigation from the info drawer, Quick Look opening, broader image thumbnails, and file-type icon previews.

Improvements

• Search and navigation upgrades: added comprehensive search indexing, bounded filename and content search, CamelCase ranking, directory / symlink handling, cancellation, and more reliable Hub / Home search behavior.
• Viewer experience upgrades: PDF and DOCX now have document outlines, PDF same-page navigation is more accurate, and PPTX, DOCX, Markdown, Office inline previews, image viewers, and media viewers received rendering, link-handling, and resource-cleanup polish.
• Release versioning upgrade: version numbers now use `major.YYYY.MDD`; this release moves from 2.x to 3.x. Same-day patch releases can add a semantic suffix after `major.YYYY.MDD`.

Fixes

• Rolled up a large set of recent review fixes for safety boundaries, including IPC / URL / file path / archive entry / local media / Webview / Router API validation, response-size limits, and cache-control handling.
• Rolled up stability and performance improvements across async file I/O, download and polling cancellation, object URL and timer cleanup, search / indexing budgets, window restore bounds, startup white-screen latency, and build fallback paths.

changelog

2.2606.1017

New Features

• Upgraded file browsing operations: File Navigator and Path Explorer now support Finder-like copy, cut, paste, drag-to-move, copy from system drops, and marquee multi-select. Selection state, context menus, and Path Explorer behavior are more consistent.
• Enhanced video timeline: generation history now supports deleting scenes, in-place redo, and preserving discarded takes. Keyframe / reference-image drop zones are easier to click, with confirmation dialogs and success / failure toasts for redo and delete actions.
• Improved admin dashboard: Overview and Analytics are merged into one Dashboard with multi-window KPIs, overall totals, device funneling, supported format / app totals, and per-user usage breakdowns on user detail pages.
• Editor and viewer upgrades: Notebook gains zoom and block gutters; the ICS viewer is renamed to Calendar Editor; archive viewing, drag scrolling, Mindmap PNG export, vector image export, and common document viewing have been improved.

Fixes

• Fixed Python / Jupyter environment installation so it can be stopped or closed reliably; cancellation now aborts requests, terminates subprocesses, and cleans up incomplete installs.
• Fixed a startup root race in File Navigator, conflicts between viewer drag scrolling and selection state, and unstable video continuation first-frame / ffmpeg path handling.
• Fixed non-OK Web API responses being treated as normal data, better-auth session read failures breaking API routes, and unclear admin analytics failure states.
• Fixed missing Windows / APPX icon assets and Vitest include globs using incompatible path separators on Windows.

Improvements

• Completed a broad i18n pass for Admin, Dashboard, Chat, native menus, and related surfaces; added .gitattributes recognition as a text format.
• Migrated in-repo agent configuration to the Codex directory and added design docs, implementation plans, and regression tests for this release cycle.

changelog

2.2606.402

New Features

• Upgraded AI video generation (Seedance 2.0 multimodal): you can now add more reference material when generating a video — multiple reference images, a reference video, and reference audio to guide the content and style; a new "Generate audio" toggle lets the video include its own voiceover / sound effects, plus a "Watermark" toggle (off by default). These controls show or hide automatically based on the capabilities of the selected model.

Fixes

• Fixed an intermittent parameter error that could cause video generation to fail on some models (including first/last-frame and keyframe scenarios), making generation more stable and reliable.

changelog

2.2606.302

New Features

• Portrait / landscape orientation for AI image & video generation: a new orientation picker (Portrait 9:16 / Landscape 16:9 / Square 1:1, portrait by default), suited to mobile short-video and other vertical scenarios; the chosen aspect ratio is passed through to the upstream model with the generation request.

Fixes

• Fixed artifact download failures for built-in AI image / video / voiceover — generated results now download and save locally as expected.
• Fixed an intermittent 503 / no available upstream error in built-in chat under certain built-in provider configurations.
• Video generation now correctly forwards the random seed, negative prompt, motion strength, and camera-motion parameters.
• Fixed a rendering (hydration) warning on the app logo in some scenarios.

Improvements

• The image / video generation toolbar is now responsive: on a narrow panel, secondary controls (orientation / count / multi-view / advanced params) collapse into a "⋯" menu, and the model selector is no longer squeezed or truncated.
• Controls available for built-in generation models (reference image / seed / negative prompt / keyframes / camera motion, etc.) now show or hide based on each model's declared capabilities.
• More accurate UI language detection: full language tags reported by the OS / browser (e.g. zh-Hans-CN, zh-TW, en-GB) are normalized to a supported UI language instead of incorrectly falling back to English.
• The new and improved UI copy above is localized in all 5 languages.

changelog

2.2606.115

New Features

• OpenFiles built-in AI service (ready right after sign-in): signing in to your OpenFiles account automatically grants an "OpenFiles built-in" provider — without configuring any vendor keys yourself, use all AI capabilities: chat, AI image generation, AI video generation, AI voiceover (TTS), and voice cloning are all built-in and login-ready; built-in OFC balance badge + one-click top-up (Stripe / Alipay / WeChat), with a prompt that guides you to top up when credits run low; the model selector shows two tracks, "OpenFiles built-in" and "My Providers".
• AI video generation (enhancements): supports the full text-to-video pipeline; supports multi-keyframe sequences (N keyframes → N−1 interpolated segments auto-stitched together) and video character consistency (keyframes lock the subject); outputs embed an AIGC compliance marker.
• AI image generation enhancements: reuse the last random seed for reproducibility; controlled multi-view of the same subject (e.g. six-view e-commerce shots); semantic reference images driving consistency (character / product / scene references); generated images embed AIGC metadata + an "AI-generated" badge on the card.
• AI voiceover / speech synthesis: text-to-speech lets you pick the voice, language, and speed with instant preview; multi-speaker dialogue scripts keep each speaker's voice / accent consistent throughout (segment-by-segment synthesis + auto-stitching); audio outputs embed AIGC compliance metadata + a badge.
• Voice cloning (voice replication): upload a voice sample to clone a custom voice, gated by a built-in authorization confirmation (per the Measures for Labeling AI-Generated Synthetic Content); cloned voices can be used directly in voiceovers.
• Consistency asset library: unify "character / product reference images" and "voiceover voices" into reusable assets with stable IDs, selectable with one click across image / video / voiceover generation — enabling cross-modal consistent reuse where "the same character keeps the same voice".
• Markdown editor: the WYSIWYG editing area adds find & replace; the rendered view adds zoom controls; pasted Markdown source is automatically parsed into rich-text nodes.

Fixes

• PPTX: filled in rendering details for master / layout background shapes, theme colors and custom / preset geometry, group transforms, directional arrow presets, gradient stop ordering, and text color schemes; "Fit to window" now works on any page.
• Mindmap: tree layout now arranges by path rather than per-level column spacing; fold icon stroke is constant and decoupled from connector depth.
• Multi-tab: global keyboard shortcuts and pagination shortcuts now act only on the currently active tab, avoiding accidental cross-tab triggers.

Improvements

• All of the AI generation and asset library UI copy above is synced across 5 languages (Simplified Chinese / English / Japanese / Korean / Traditional Chinese).

changelog