TidyAI：讓 AI 替你整理檔案，讓你輕鬆休息

Have you ever opened your Downloads folder and found hundreds of files scattered everywhere?

PDF documents, screenshots, college notes, images, ZIP archives, code files, certificates, and random documents — all mixed together in one place.

You tell yourself, "I'll organize these files later." But later never comes!

Manually sorting files is not difficult, but it is repetitive, time-consuming, and easy to postpone. As developers and students, we should spend more time learning, building, and creating rather than organizing files one by one.

That's the idea behind TidyAI — a privacy-focused, AI-powered file organization assistant designed to make everyday file management easier.

Instead of spending time sorting files manually, imagine having an assistant that helps you understand your files, suggests where they belong, and handles repetitive organization tasks with minimal effort.

The philosophy is simple:

Work smarter, automate the boring stuff, and let AI handle the repetitive tasks. 🦥

💡 What I Built

TidyAI is designed to turn a messy collection of files into a structured and organized workspace.

The project combines regular file-system automation with an open-weight AI model. Traditional file organizers usually depend on file extensions or predefined rules. TidyAI aims to go further by using AI to help interpret file context and recommend meaningful categories when simple rules are not enough.

For example, a folder might contain:

python_notes.pdf
IMG_20261020.png
resume_final_v3.pdf
certificate_cloud_computing.pdf
project_backup.zip
lecture_recording.mp4

Instead of manually reviewing every file, users could receive organization suggestions such as:

Documents: PDFs, Word files, and text documents
Images: Screenshots, photographs, and graphics
Study Materials: Notes, assignments, and learning resources
Certificates: Course certificates and achievements
Archives: ZIP files and compressed folders
Videos: Recordings, tutorials, and other video files

These categories are examples of the intended experience. More specific classifications, such as distinguishing a certificate from ordinary study material, may require inspecting document content rather than relying only on filenames or extensions.

The goal is to make file organization more intelligent, transparent, and convenient without taking control away from the user.

😫 The Problem: Digital Clutter Is Real

Our computers gradually become storage spaces for files we no longer organize.

This creates several everyday problems:

1. Finding files takes longer than expected.

You remember downloading a document, but you cannot remember its exact filename or location.

2. Downloads folders become overwhelming.

Every new download adds another file to an already disorganized collection.

3. Manual organization is repetitive.

Creating folders, checking extensions, renaming files, and moving them individually wastes time that could be spent on more valuable work.

4. Existing automation can be too basic.

Rules based only on file extensions cannot always identify what a document is actually about. Two PDF files might represent completely different things — a college assignment and a course certificate, for example.

5. Privacy matters.

Personal documents can contain sensitive information. Uploading every file to an external AI service just to organize a folder may not be desirable.

TidyAI explores how local AI and ordinary automation can work together to address these challenges.

✨ Key Features

1. Smart File Categorization

TidyAI is designed to use file information to recommend meaningful categories.

Basic classification can use filenames, extensions, and metadata. When a file's purpose is unclear, an AI model can help interpret available context, subject to the information provided to it.

The objective is to reduce the need to inspect every file manually.

2. Automatic Folder Organization

Once categories have been suggested, the application can help organize files into appropriate folders.

For example:

Downloads/ → Documents/

Downloads/ → Images/

Downloads/ → Study Materials/

A safe workflow should let users review proposed changes before applying them. This helps prevent accidental moves, duplicate filenames, and unexpected changes to existing files.

3. Local AI Processing

One of the main ideas behind TidyAI is to explore running an open-weight model locally rather than depending entirely on a hosted AI API.

With a compatible local model and runtime, AI inference can take place on the user's computer. This can reduce the need to send file information to external services.

Local processing still depends on the actual implementation: file operations, telemetry, downloads, or optional integrations must also be considered when evaluating privacy.

4. Simple and Beginner-Friendly Interface

File management tools should not require users to understand complicated commands.

A simple interface can make it easier to select a folder, preview suggested categories, review proposed actions, and organize files.

The focus is on making the workflow understandable for students, developers, professionals, and everyday computer users.

5. User-Controlled File Operations

Automation should not mean losing control.

A responsible file organizer should provide a preview of proposed changes, allow users to confirm them, and handle filename conflicts safely.

The intention is to make organization convenient without silently moving or overwriting important files.

⚙️ How It Works

The proposed workflow combines deterministic file handling with AI-assisted classification.

Step 1: Select a folder

The user chooses a directory they want to organize, such as Downloads or a project folder.

Step 2: Scan the files

Python inspects supported files and collects relevant information, such as filenames, extensions, and available metadata.

Step 3: Analyze and categorize

The application applies straightforward rules first. When additional context is useful, a locally running open-weight model can suggest a category based on the available information.

Step 4: Preview the results

The user reviews the proposed categories and destination folders before any changes are made.

Step 5: Organize the files

After confirmation, the application moves supported files into their destination folders while checking for conflicts and handling errors.

Step 6: Review the outcome

The application can display a summary of completed operations and any files that could not be organized.

This hybrid approach is important. AI is useful for interpreting ambiguous information, but predictable operations such as checking file extensions and moving files should remain under explicit programmatic control.

🛠️ How I Built It

The project is planned around Python-based automation, an open-weight AI model, local inference, and a lightweight user interface.

Technology Stack

Python: Core application logic, file scanning, categorization rules, and file operations.
Gemma: An open-weight model option for AI-assisted categorization and interpretation.
Ollama: A local model runtime for running a compatible Gemma model.
Streamlit: A lightweight interface for interacting with the file organizer.

The final model version and configuration should be selected according to the computer's available memory, performance requirements, and the capabilities of the chosen model.

Why Combine AI with Traditional Programming?

Not every task needs a language model.

Identifying a .png file as an image is a straightforward programming task. Moving a file into a folder is also something Python can handle reliably.

However, understanding that semester3_os_notes.pdf is probably related to operating systems or that a document might be a course certificate requires more contextual interpretation.

TidyAI's approach is to use ordinary code for predictable tasks and AI where contextual assistance can add value.

This keeps the workflow easier to understand and avoids using AI unnecessarily for simple operations.

🔓 Why Does Open Innovation Matter?

Open innovation is particularly valuable when building tools that people may want to customize, inspect, and run on their own machines.

For TidyAI, the choice of an open-weight model is not just about adding AI to an application. It is about exploring a different way to build and use AI-powered software.

1. More Control Over the AI Workflow

A locally available model gives developers the opportunity to experiment with prompts, supported model versions, and inference configurations without depending entirely on a proprietary hosted inference endpoint.

2. Privacy-Focused Design

A local model can process the information supplied to it without requiring that information to be sent to an external model provider.

For a file organizer, this is an important design advantage. However, local inference alone does not guarantee complete privacy; the rest of the application must be designed accordingly.

3. Reduced Dependence on Paid APIs

Local inference can reduce usage-based API costs for supported workloads. Users still need compatible hardware, storage, and electricity, and some models may require substantial system resources.

4. Accessibility for Developers

Open-weight models make it easier for developers and students to experiment with AI-powered applications, learn how inference works, and customize their tools.

They also create opportunities for contributors to improve categorization, usability, error handling, and documentation.

5. Learning Through Building

For me, this project is also an opportunity to explore the practical side of AI engineering: choosing a model, integrating it with a Python application, combining AI with deterministic logic, and designing a workflow that people can actually use.

A closed API can be a useful choice for many applications, but open-weight models offer an alternative when local execution, customization, and greater control over deployment are priorities.

🔐 Privacy and Safety Considerations

Because TidyAI is intended to work with users' personal files, safety should be part of the design rather than an afterthought.

The application should aim to:

Process files locally when using the local inference configuration.
Avoid sending full document contents to the model when filenames or metadata are sufficient.
Ask for confirmation before moving files.
Avoid overwriting existing files without explicit permission.
Handle inaccessible files and unsupported formats gracefully.
Exclude sensitive system directories from automatic organization by default.
Provide clear feedback about what was changed.

These safeguards can help make automation more predictable and give users confidence in the process.

🚀 Future Improvements

There are several directions in which TidyAI could evolve.

Duplicate File Detection: Identify possible duplicate files to help users reclaim storage space.

Custom Categories: Allow users to define categories and organization rules that match their own workflow.

Content-Aware Organization: Explore supported document text extraction so the model can classify files using their contents when necessary.

Undo and Recovery: Maintain an operation history and support safe reversal of file moves.

Batch Processing: Organize larger folders efficiently while reporting progress and errors.

Custom Naming Suggestions: Suggest more descriptive filenames while preserving the original files until users approve the changes.

Cross-Platform Support: Improve compatibility across macOS, Windows, and Linux.

These are potential improvements, not claims that every feature is already implemented.

🎯 Who Can Benefit from TidyAI?

TidyAI is intended for people who regularly download, create, or manage large numbers of files.

Students can keep assignments, notes, and learning resources organized.
Developers can reduce clutter in folders containing documentation, screenshots, and downloaded tools.
Content creators can organize images, recordings, and project assets.
Professionals can manage documents and downloads more systematically.
Everyday computer users can spend less time sorting files manually.

The underlying idea is simple: small repetitive tasks add up, and automating them can make everyday computer use more convenient.

🎥 Demo

Screenshots:

The demo should show the folder selection process, categorization suggestions, preview of proposed changes, and the final organization result.

🌱 What I Learned

Building TidyAI is an opportunity to explore a practical question: how can open-weight AI make an ordinary, repetitive computer task easier?

It also highlights an important lesson about AI development. A useful AI application does not always need a complicated agent or a large collection of tools. Sometimes, combining a suitable model with simple, reliable programming is enough to create a valuable experience.

The challenge is to use AI where it genuinely helps, keep important operations predictable, and make the final tool easy to use.

🦥 Final Thoughts

We often spend too much time doing small tasks that could be automated.

Organizing files is one of them.

TidyAI explores how open-weight AI, local inference, and simple automation can help people spend less time managing digital clutter and more time doing what matters to them.

Because productivity is not always about working harder. Sometimes, it is about making repetitive work disappear.

Less sorting. Less effort. More time to create. 🚀

If you are interested in open-source AI, local models, or practical automation, I'd love to hear your thoughts and suggestions.

What repetitive task would you automate with an open-weight AI model?