SlothSort：以本機 AI 整理檔案，並由用戶掌控操作

What if your AI could organize your files without requiring you to hand your files over to a cloud AI service?

Meet SlothSort — a privacy-first Windows desktop application that uses local AI to propose file organization, while keeping the final decision in your hands.

Built for the Hacktoberfest Open-Source AI Challenge — Week 3: Slow and Steady.

📂 The Problem: Digital Clutter Never Sleeps

Downloads pile up. Screenshots multiply. Documents disappear into folders named New Folder, New Folder (2), and final_final_REAL.

Manually sorting everything is tedious. Sending file information to a cloud AI service isn't always a trade-off users want to make.

I wanted to explore a different approach:

What if AI handled the tedious proposing, but never took control away from the user?

That idea became SlothSort.

🎬 See It in Action

Watch the SlothSort v0.1.1 demo:

In this demo, you'll see the complete journey:

🧠 Local AI proposes categories, destinations, and filenames.
👀 The user reviews the suggestions before anything changes.
✏️ Proposed changes can be edited before approval.
✅ Only explicitly approved changes are applied.
↩️ Operation history and undo provide a way to review and reverse changes.

This isn't just a chatbot recommending folders. It's a desktop application that connects AI suggestions to actual file operations—with a deliberate human approval step.

🛠️ How I Built It

SlothSort combines a modern desktop stack with a locally running open-weight model.

The stack

Electron — desktop application and controlled access to filesystem operations.
React 19 + Vite + TypeScript — responsive user interface and typed application logic.
Ollama — local model runtime.
Gemma 3 4B (gemma3:4b) — the open-weight model powering file-organization suggestions.

The inference workflow uses a local Ollama endpoint at 127.0.0.1:11434. After the model and runtime are installed, the core AI workflow does not require a cloud AI API.

The architecture is intentionally cautious

AI can make mistakes. A plausible filename or category doesn't automatically mean a file should be moved.

SlothSort separates suggestion from execution:

The scanner gathers the context needed for organization.
Local AI generates proposed changes.
The user reviews and can edit the proposals.
Approval is validated before the operation is executed.
The application records operations so users can inspect history and undo supported changes.

The Electron renderer runs with contextIsolation enabled and nodeIntegration disabled. Privileged filesystem operations stay in the main process behind a narrow, typed IPC interface.

I also designed the workflow without a file-deletion operation. The application focuses on moving and renaming files, with the user remaining in control.

🔓 Why Does Open Innovation Matter?

For a file organizer, privacy isn't an abstract feature. Files can contain personal documents, work material, screenshots, and information users simply don't want sent to an external AI service.

Using an open-weight model locally makes a different design possible.

1. Your AI can run on your machine.

With Ollama and Gemma installed, SlothSort can generate suggestions locally instead of depending on a hosted AI inference API.

2. The implementation can be inspected.

The source code is available for developers to examine how the AI workflow, approval process, and filesystem operations are implemented.

3. The user retains control.

Local inference is only one part of the design. Review, explicit approval, operation validation, and undo help prevent AI suggestions from becoming unchecked filesystem changes.

4. Developers can build on the work.

The combination of open-source tooling and an open-weight model gives other developers a foundation they can inspect, reproduce, adapt, and improve.

Open innovation matters here because it lets us explore not only what AI can automate, but also how to build that automation with more user control.

🦥 The Lazy Payoff

The goal isn't to make people stop caring about their files. It's to make organization require less repetitive decision-making.

Instead of choosing a destination and filename for every file from scratch, users can review AI-generated proposals and approve the changes they actually want.

I have put the project through automated testing and packaged Windows release validation. The repository also documents the architecture, setup instructions, testing evidence, and known limitations.

I won't claim a made-up percentage of time saved. A reliable time-saving figure would require a repeatable comparison against manual organization. For now, the measurable product goal is straightforward: reduce repetitive sorting work without removing the user's control over their files.

Screenshots

The following screenshots are tracked in the repository and come from the packaged Windows validation workflow.

Launch and scan	Review proposals

Approval and execution	History and undo

🚀 Try It Yourself

Source code:

SlothSort 🦥

Local AI suggestions for safer file organization. SlothSort is a Windows desktop application that scans a selected folder, proposes categories, filenames, and destinations through local Ollama inference, and waits for your approval before moving anything.

15-second overview: SlothSort turns a messy folder into a reviewable queue of file-organization proposals. The model runs through Ollama on the local machine, proposed moves are validated in the main process, and approved operations are recorded so supported moves can be undone.

Start here

Repository: github.com/Pranjal-debug/slothsort
Windows release: SlothSort v0.1.1
Direct ZIP download: SlothSort-v0.1.1-win-x64.zip
License: MIT

Demo status: The concise 1080p demo recording is prepared locally at artifacts/day-2-audit/voiceover/SlothSort-v0.1.1-demo-video.mp4, but it is not currently a tracked repository or public release asset. The README therefore uses the tracked screenshots below instead of linking an untracked binary.

Why SlothSort?

Sorting files is easy to postpone and difficult to do safely in bulk. A plain AI-generated…

The repository README contains the installation instructions, architecture details, and development workflow.

The release is for Windows x64. Because the application is unsigned, Windows SmartScreen may display a warning.

🏆 Prize Category

Best Use of Gemma

SlothSort uses Google's open-weight Gemma 3 4B model through Ollama as its local AI engine.

Final Thought

I built SlothSort around a simple principle:

Let AI suggest. Let humans decide. Keep the workflow local.

AI doesn't need unrestricted access or the final say to be useful. Sometimes, the best automation is the kind that saves you effort while leaving you firmly in control.

Thanks for checking out SlothSort. If you explore the code, find a weakness, or have an idea for making local AI file organization better, I'd love to hear your feedback.

🦥 Built for the Hacktoberfest Open-Source AI Challenge — Week 3: Slow and Steady.