Continuous evolution, just for a smarter experience

Newly added quick naming mode for file analysis, making file analysis faster

Eliminate the problem that small models sometimes reuse original file names

Use anydoc to replace markitdown and other tools for reconstruction, greatly improving performance and occupying less resources.

Qwen3.8-27B, Bonsai-27b-1bit and high-speed model LFM2.5 series

Dynamically adjust AI engine configuration parameters based on user hardware capabilities, and the engine runs in the best mode

The interface adapts to the length of multi-language content, and AI replies accurately follow language requirements.

Supports OCR text recognition, content extraction, file type recognition, metadata extraction, and cover generation for any file

During batch analysis, you can choose whether to skip analyzed files.

The saved virtual directory supports adding newly analyzed files

The saved virtual directory supports continued sorting of uncategorized files.

The AI-generated folder directory tree can be edited and modified at any time during the whole process to customize the exclusive directory structure.

Load the native high-definition icons of the file system to make the files more recognizable

The three types of file list views support free scaling and can flexibly adapt to the screen display space.

When OCR is turned off, the document parsing speed increases by 2‑3 times, and remains unchanged when it is turned on.

File information extraction and AI analysis tasks run in parallel, increasing document analysis efficiency by 10‑20%.

The analysis options are simplified into two modes: simple analysis and AI analysis.

Supports search for all information of analyzed files

Further constrain the smart file name format generated by AI to avoid small models not following

Optimized analysis time-consuming statistical chart, clear and intuitive data structure

Adapted to Nanbeige 4.2 3B, running with minimum 3G VRAM, smarter than 9B model

Added a new local conversation entrance to chat privately with jailbroken models offline

Supports fast and high-precision mode switching, and the text recognition accuracy rate increases from 60% to 90%

Smart name/original file name can switch primary and secondary display

Switch to control whether to display deleted and migrated files, file recovery, and automatic correlation analysis data

Automatically skip system and program installation directories, etc., to reduce time-consuming useless scanning

Improved file classification accuracy and plan generation success rate

The time-consuming indicators of file analysis are presented graphically, allowing you to weigh configurations.

The performance of the classification dimension tree has been optimized. Now there is no lag when selecting all or inverting the selection.

Customize the scheme to accurately generate the directory structure you enter

The free version's private directory quota is increased from 200 to 500; reminders are only displayed when quotas are tight

Supports minimizing to the system tray, and switching working directories via right-click menu in the tray

File analysis queue changed to column display, not affecting the main operation workflow

File analysis queue supports display in a new window, saving more space

Visual optimization of multiple settings, understood at a glance

Tag tree adds intersection and union switching, filter files as you wish

Since 3.0 release, accumulated multiple bug fixes

Preview for 200+ formats, including Office, PDF, e-books, compressed archives, 3D assets, etc.

Automatically verifies and corrects actual file extensions; files without extensions can be previewed and opened directly.

AI generates multiple file organization plans with one click; new guided step-by-step organization process.

New support for multiple virtual directories to meet the needs of multi-perspective, multi-dimensional file categorization and management.

Intuitive process navigation bar, organizing steps at a glance

Expanded file dimension system, added pre-classification mechanism for more precise file categorization

Fine-tuned AI label generation strategy, allowing low-spec small models to output high-quality labels

Supports manual deletion of tags for more flexible management

Provides three selectable analysis modes, significantly improving the speed of full-volume file scanning and analysis

OCR support for PDFs, Office documents, and images

Enable or disable AI deep thinking mode freely to balance response speed and analytical accuracy

All functional blocks support drag-and-drop width adjustment, with automatic memory of layout and one-click reset

Added user levels and Firefly Points consumption system to unlock more exclusive benefits

Detailed iteration and optimization of interface elements, adaptive to different screen widths, with comprehensive upgrade of operation feel

OpenSkill (to be listed), OpenClaw and other external AI tools can interconnect with Firefly AI Folder

Batch fixes for known program defects, significantly improving software running fluency and stability

Optimize the AI ​​engine installation logic and significantly reduce the probability of engine deployment failure

Added AI engine CPU operating mode switch to make hardware adaptation more flexible

Newly added lightweight high-speed model Gemma-4-E2B-it-qat-MTP

Unified adaptation to PDF, Office documents, and compressed packages to complete integrated extraction of multi-format file content and metadata

Office document parsing content is uniformly output in Markdown standard format

Added minimum unit function switch, turned off by default

The stable version installation package is offline and the entire domain is upgraded to the all-in-one version.

Fixed multiple known program issues and comprehensively improved software operation stability

Simplified setup process with automatic GPU driver matching for out-of-box deployment

Display activated model details for clearer configuration management

Navigate and select files via arrow keys in the item list for improved efficiency

Optimized interface layout for search, batch analysis and other core modules

Added breadcrumb path navigation for quick directory switching

Restyled into tabbed layout for organized and intuitive data viewing

Improved AI reasoning logic to deliver focused, non-divergent analysis results

Resolved multiple known glitches for enhanced program stability

Added Ali Moda community source to completely solve the problem of some users being unable to download models.

Add many service providers at home and abroad, and provide Qingxi’s free instructions and usage information

Graphics card driver needs to be upgraded detection

Solve the bug that the cloud mode setting page cannot be switched

Upgrade the application base to the latest

Enhance stability and improve running speed

A new "stable version" of the built-in engine is available for download

Natively adapted to NVIDIA graphics cards, fully compatible with all AMD, Intel, and Apple series core graphics and independent graphics

Keep the local built-in engine, add Ollama cloud access, and deploy various local large models at will

Qwen3.6 MTP model is now available for high-performance graphics card users to try out

Online right-click shortcut menu, one-click direct access to the real storage directory of files

Optimize multi-selection operation, support mouse frame selection and Shift/Ctrl quick batch selection

A new waterfall flow browsing view has been added, and the look and feel of image and text previews has been greatly upgraded.

Built-in AI engine and model, no additional configuration required, ready to use out of the box

Optimize local reasoning logic and increase local AI response speed by 30%-100%

Compatible with core graphics, integrated graphics and devices without independent graphics, all models support local AI functions

Upgraded tag tree system, multi-select tags can filter and display corresponding files in real time

Optimize massive file scenarios and greatly improve virtual directory loading speed

Reconstruct the UI layout, refine and optimize interaction details, and make operations smoother

Added shortcut operations such as the backspace key and Alt+direction keys to improve operating efficiency

Launched enterprise authorization service, supporting one-click deployment and offline localization use

Optimize the opening and loading speed in the scenario of massive analyzed files

Fixed the abnormal problem of failure to download the model on the settings page

When the test connection fails, the detailed failure reason is displayed.

Streamlined file thumbnail card style, more concise visual appearance

Fixed an issue where some list columns were not adapted to dark mode

Fix tag counting errors, search failures, missing tag generation, etc.

A new Zhihu circle entrance has been added in the lower right corner to facilitate feedback and communication.

Supplement and improve novice guidance and operation tips

Add Alibaba Magic Community sources to solve the problem of slow model downloads

Automatically switch domestic and foreign lines to solve the problem of installation failure

Support CPU mode to use minimal lightweight model

Data security protection upgrade

Added Gemma4 model to support recognition and analysis of spoken language recording content

Open for use, basic quota is free; invite friends to unlock unlimited quota

The new invitation code mode makes unlocking rights more ceremonial.

Massive files support fast positioning of keywords with more than three words

Open a super large directory in seconds and say goodbye to lags

The theme color scheme reduces saturation and the interface is fresher and eye-friendly.

Fixed an issue where directory monitoring could not automatically analyze newly added files

Fix the problem of custom filtering rules not taking effect

Fix theme label parent classification error

Supports dynamic adjustment of analysis content length based on the model, and cloud models have longer prompt words

Supports quick sorting of selected files and export of virtual directories

Supports AI intelligent analysis and understanding of multimedia types such as text and images

Based on file link technology, personalized directory organization can be achieved without taking up additional space

AI-driven intelligent classification and renaming greatly improves file management efficiency

Provides rich directory tree labels and flexible dimension management system

The core supports offline operation of local large models to ensure that data does not leave your device

AI-driven document quality assessment to help quickly identify high-quality content

Perfectly compatible with Windows, macOS and Linux operating systems

Supports rapid positioning of files by content summary, tags, quality and other multi-dimensional dimensions