ToolKnit Desktop 3.2

One toolbox for the little tasks scattered across your desktop.

73 desktop tools · 12 categories · 77 CLI / MCP capabilities · Local first

Source licenses and separate PDF-to-Word authorization

About the project and me

Hi, I am Zihang Dong, an independent developer turning everyday annoyances into small tools. ToolKnit is a spider weaving a growing web. Each need adds a thread and becomes a tool. People discover it, use it and help improve it. The boy is me; the spider on his head is the toolbox's little companion.

Try the web app first

Open ToolKnit.com in a modern desktop or mobile browser. No installation needed. 95 web tools cover PDF, images, audio, video, text, calculators, time, creative tasks, AI and self-tests. Most need no account. Supported file tasks run locally in your browser. Network requirements are listed in each tool.

Then bring it to your desktop

For everyday file work, install the Windows app. V3.2 contains 73 tools, with PDF to Word, Images to PDF, expanded floating history and saved phrases. Set an output folder and favorite your tools. Install models only when needed. Enable one-click recording mode for startup, clipboard monitoring and floating history. Press Alt + Q to capture. The home page's V3.2 feature filter groups six new or enhanced tools. See Get started for installation requirements and instructions.

People weaving it together

Zihang Dong Product design · Development · Maintenance

Joshua-Zion Color spaces · Interaction improvements

hqyx2025 Project verification · Bug testing

Thanks to everyone who uses, tests and improves ToolKnit. Joshua-Zion's contributions: PR #23, PR #67.

Contents

01 The idea · 02 Recording mode · 03 Screenshots · 04 Clipboard

05 Files · 06 Pictures and sound · 07 CLI / Agents · Actual screenshots

08 Get started · Tool catalog · Support and feedback · Run and build from source · Technology stack · License · Star History

These are referral links. Pricing, model capabilities and availability are determined by the providers.

01 / Fewer apps, more things done

↑ Back to contents

Start with one small tool, then weave more of the desktop's everyday friction into one web.

Crop a PDF page. Find the text you just copied. Annotate a screenshot. Remove an image background. Small jobs often mean switching between several applications.

ToolKnit brings PDF, PPT, image, audio, video, text, system and developer tools together. Use the desktop app for local files, the web app for quick access, and CLI or Agents for suitable automated workflows.

02 / Keep the things you want to remember

↑ Back to contents

Enable one mode in Settings to bring three features together.

Recording is opt-in and can be paused. Turning the mode off restores previous settings. Each switch also works independently. Dismiss the home prompt for this session or permanently. The Settings switch remains available. No keystroke or browsing logging. Monitoring stops when the application fully exits.

03 / A screenshot is just the start

↑ Back to contents

Press Alt + Q by default to capture a window or region. You can change or disable the shortcut.

Add shapes, arrows, pen strokes, mosaic and text. Undo, redo or hold to compare with the original. Text handles fit the content. The text editor has a quick-copy button. Copy, save PNG / JPG or open the pinned viewer. Reveal a saved file in its folder and exit capture mode. Find the tool under Creative. Screenshots share history with content copied from other apps; older images can be reopened.

Screenshot translation supports Chinese and English only. Without AI configuration, local OCR feeds the local translator. With image-capable AI configured, OCR locates text and AI corrects it before local translation. If AI fails, you choose whether to use OCR instead. The full translation remains copyable even when it cannot fit.

Translation components are downloaded on demand: approximately 192 MiB downloaded / 225 MiB installed, separate from the main installer. External AI latency depends on the provider. There is no fixed completion-time guarantee. Check recognition and translation results.

04 / Find what you copied

↑ Back to contents

The note you copied a moment ago does not have to disappear.

Text, images and file paths stay in a searchable timeline. Filter, favorite and store records with local encryption. Monitoring preferences survive restarts.

The recording switch sits at the bottom with a slim scroll indicator. Off-screen windows slide back into view. Click the menu button again to collapse it. Opening history restores the main window. Copy feedback appears at the primary screen's bottom without taking focus. Drop support depends on the receiving app. Dragging never sends a message automatically. Saved phrases survive history clearing and rotation.

Only newly copied content is recorded. Windows Win + V history is not imported. Pause or clear sensitive records as needed. A Windows-account-protected backup can restore a missing key file. Without a usable key, the old database is preserved; lost encrypted content is not guaranteed to be recoverable.

05 / Keep file work in one place

↑ Back to contents

Move from preview to output with fewer detours.

PDF to Word: convert, then edit

Local analysis reconstructs text, paragraphs, tables and columns into editable DOCX. No Python, Office, GPU or extra model needed. Handles papers, contracts, ebooks, invoices, quotes and common tables. Improves Chinese fonts, reading order, spacing and pagination.

Three panes: settings, preview and page navigation. Select pages, cancel processing or export a fresh result. Fixes include table backgrounds blocking text, short fields becoming images and reader-specific spacing. Complex formulas and uncertain regions remain local images. No scan OCR. Not every element will be editable or render identically in every reader. Limit: 64 MiB / 100 pages per file. Only enabled AI processing sends necessary regional images and text. This custom engine has a separate source license. Commercial and noncommercial reuse both require written permission. Read the terms.

Images to PDF: turn pictures into a document

• JPG, PNG, BMP, WebP, GIF and common TIFF files. Multi-page TIFF expands; animated files use the first frame.
• Drag to reorder, rotate or remove. Choose A4, original dimensions, margins and orientation.
• Original color, grayscale or scanned-style output, with three quality levels. Transparency becomes white.
• Up to 100 pages, 64 MiB per image. Processes one page at a time, with cancellation, retries and unique output names.

PDF editing and selection: start with a blank page

Create a blank A4 PDF without uploading a file. Add multiline text, images and shapes, with undo and redo. Improves rotated-page editing, edge placement and preview/export consistency. Fixes missing images after undo and stale state on rapid reopen.

The split workspace starts with no pages selected. Enter 1-10, then 15-20, to select 16 pages without 11-14. Select all, enter a single page or overlap ranges without duplicates. Multiple files use continuous numbering. Each addition clears the input for the next range.

Zoom PDF previews around the pointer with the wheel. Pan with the middle button or Space. Sidebars and text areas still scroll. Shared navigation, themes, input underlines and completion dialogs reduce flashing and clipped controls.

Output and cleanup

Excel's All worksheets option includes hidden sheets and changes only a temporary copy. Text-layer PDF extraction needs no AI key. Use the separate AI vision tools for scanned text. Duplicates are checked by content. Nothing is selected for deletion by default. Keep at least one copy per group; confirmed files are rechecked before moving to the Recycle Bin.

File tools preserve originals by default and write results to the output directory. Visual covering in the PDF editor is not secure redaction, and a scanned-style PDF does not prevent copying. Cleanup safeguards cannot replace your judgment about your files.

06 / A workbench for pictures and sound

↑ Back to contents

Crop, compress, convert and stitch images; generate icons or remove backgrounds with a local model. Convert audio and video, trim audio, extract sound, capture video frames, create GIFs, transcribe offline and edit Markdown.

FFmpeg, LibreOffice, transcription and background-removal models are installed on demand. Existing usable runtimes are reused where supported. Downloads support source fallback, resume and integrity checks. Failed updates preserve usable components. Network speeds vary. Background removal offers CPU mode. Automatic mode retries on CPU after GPU failure. Missing components get a specific error. WPS / Microsoft Office is not LibreOffice and cannot replace workflows that need soffice.

07 / Work by hand, or hand it to an Agent

↑ Back to contents

Desktop for visual interaction. CLI for batches. Agents for orchestration.

The V3.2 source expands from 46 to 77 CLI / MCP capabilities, exposed through @toolknit/cli.

Desktop V3.2 and npm CLI / MCP 3.2.0 are both released. Install each independently.

Progress, cancellation and explicit errors are supported. Originals and existing outputs stay protected. Unicode paths, long paths and model downloads are improved. Windows, clipboard monitoring and system cleanup remain desktop interactions. 77 interfaces do not map one-to-one to 73 desktop tools.

Use @toolknit/cli 3.2.0 or later, independently of the desktop app. See the complete catalog for all 77 capabilities. Requires Node.js 22.13+.

npm install --global @toolknit/cli
toolknit doctor --json
toolknit --help

{
  "mcpServers": {
    "toolknit": {
      "command": "toolknit",
      "args": ["mcp", "serve"]
    }
  }
}

The actual workspaces

Windows desktop

An actual capture of this repository's 3.1 home interface, rendered in a browser and converted to grayscale. Native file, shortcut and system features require the installed Windows app.

No-install web app

An actual capture of ToolKnit.com, converted to grayscale. Desktop and web tool scopes, dependencies and offline support follow each product's own documentation.

The comics share the approved Chinese-language artwork in both READMEs. Their accessible descriptions and the surrounding English text explain each scene and feature.

08 / Get started

↑ Back to contents

Requires Windows 10 1803 / build 17134 or later, or Windows 11, and Microsoft Edge WebView2 Runtime. The installer includes a bootstrapper. A missing WebView2 runtime needs an online download. Restart before using the app when requested. Windows 7 / 8 / 8.1 are unsupported. Available installer versions follow the download page. See the code signing policy for signing status.

Installation and everyday use

Update checks read GitHub and AtomGit. Downloads go to the official website. See the changelog for details.

Local first, with clear boundaries

Ordinary file tasks run locally by default. When you actively use an AI feature, necessary file content or screenshot regions are sent to your configured provider. Dependency downloads and update checks need a connection; installed offline capabilities do not require an AI key. File-size, page, pixel and batch limits vary by tool; support is not unlimited for every format.

Complete tool catalog

Support and feedback

Thanks to AtomGit for code hosting and developer collaboration. Repository mirror: AtomGit / ToolKnit Desktop.

ToolKnit recommends: help open-source ideas reach more people. AtomGit brings code hosting, project management and developer collaboration together. Discover useful tools, share your work and build good ideas with other developers. Explore AtomGit.

Report a bug · Suggest a feature · Contribution guide

If ToolKnit saves you time, a Star or one-time donation helps support development, AI testing and maintenance. Thank you to our supporters. Donations do not purchase a feature or promise implementation. Custom tools: WeChat wunanvip6. Code licensing requests, including commercial licensing: [email protected].

Run and build from source

↑ Back to contents

Prerequisites

Confirm authorization before running the source. This branch includes the PDF-to-Word Engine. Running, testing or building the complete app with that Engine requires prior written permission. Reading the source does not. Normal use of official apps and the official CLI follows their applicable terms. See the Engine license.

Use a supported Windows 10 / 11 system with Node.js 22.13+ (24 recommended), Rust stable 1.90+, Visual Studio C++ Build Tools (Desktop development with C++ and Windows SDK), and Microsoft Edge WebView2 Runtime. Initial npm and Rust dependency installation requires a connection. Install using the existing lockfiles.

Start the desktop app

git clone --branch main https://github.com/ZihangDong/toolknit-desktop.git
Set-Location toolknit-desktop\toolknit-desktop
npm ci
npm run tauri dev

This checks out main. Once authorized, run npm run dev for a frontend preview at http://localhost:1420/ by default. Native files, the system tray and global screenshots need the Tauri desktop environment.

Build an installer

From the nested toolknit-desktop/ directory:

npm run build:release-package

Builds the frontend, then the desktop app. Production settings disable DevTools. Code signing needs separate configuration. Default NSIS output: src-tauri/target/release/bundle/nsis/. A custom Cargo target directory changes that location. FFmpeg, LibreOffice and optional models are installed on demand. Source code does not include user keys or complete model packages.

Source layout

Repository root/
├── README.md / README_EN.md   Project introduction
├── BUILD.md                  Full build guide
├── assets/                   Documentation and artwork
└── toolknit-desktop/          npm / Tauri working directory
    ├── src/features/         Tool pages, logic and styles
    ├── src/app/              App orchestration and lifecycle
    ├── src/shared/           Shared UI and runtimes
    ├── src/platform/         Browser and native adapters
    ├── src-tauri/            Rust native capabilities and configuration
    ├── cli/                  CLI / MCP interfaces
    └── scripts/              Checks, tests and build scripts

Read more: Build guide · Contributing · Architecture and maintenance (Chinese) · CLI / MCP.

Technology stack

Third-party components retain their own licenses. See package-lock.json and src-tauri/Cargo.lock in the desktop project for exact versions. Optional runtimes and models are not all bundled in the main installer.

Source license

Visible source is not blanket permission to reuse it. Licenses differ by component.

Separate PDF-to-Word Engine terms

Applies to author-owned Engine material in:

• All of toolknit-desktop/src/features/pdf-to-word/.
• toolknit-desktop/src-tauri/src/native_runtime/pdf/docx_write.rs.

Engine copies and adapters in the official npm package retain the same license. Moving, renaming or extracting code does not change its terms. The official-app exception does not permit Engine extraction, reuse or building modified versions. Your input files and converted output do not become the author's property or inherit the Engine source license.

Contact [email protected]. Describe the intended use, deployment and distribution scope. An email request, donation, payment or publicly available source is not authorization.

Copyright remains with the respective authors. These are source-available licenses, not OSI-approved open-source licenses. Independent implementations of general ideas that do not copy protected expression remain unaffected. So do express hosting-platform rights and non-waivable legal rights. ToolKnit branding, domains and hosted services are not licensed by the source license. This is a summary. See NOTICE, the Engine license and Apache-2.0 for the applicable terms.

Star History