FiloAI

This is a project write-up inspired by the Hacktoberfest Open-Source AI Challenge: Week 3 — Slow and Steady.

What I Built

Have you ever opened your Downloads folder and found hundreds of files you no longer remember downloading? Documents, images, duplicate files, and random files accumulate until organizing everything feels like a project of its own.

That's why I built filoAI — a privacy-first AI-powered digital decluttering agent.

Its tagline is simple: Less digital clutter. More peace of mind.

filoAI helps users understand and organize their files without handing control of their filesystem to an autonomous AI. Instead of manually checking every file, users can scan a folder, identify duplicates, receive organization recommendations, and review proposed file moves before approving them.

The project includes:

• Folder scanning: Inspect file metadata within a configured, approved directory.
• Duplicate detection: Identify duplicate files using streaming SHA-256 hashing.
• Local AI recommendations: Use Ollama with a configurable Gemma model to suggest sensible file organization.
• User-approved organization: Preview proposed moves before executing them.
• Undo and operation history: Review completed operations and reverse supported moves.
• Recovery journal: Record file-operation states and flag interrupted operations for manual recovery.

The goal isn't to let AI take over your computer. It's to reduce repetitive work while keeping the user in control.

Demo

Project repository: https://github.com/VipinKumar04/FiloAI

The demo should show a folder scan, duplicate detection, AI recommendations, a user-approved file move, operation history, and undo.

The application is designed to run locally at http://localhost:3000. The local AI workflow depends on Ollama and the configured model being installed and running.

Code

Explore the source code on GitHub:

https://github.com/VipinKumar04/FiloAI

The repository includes the Express application, scanning and duplicate-detection services, local AI recommendations, organization and recovery workflows, automated tests, and project documentation.

How I Built It

I built filoAI with a focus on local processing, explicit user approval, and safe filesystem operations.

Technology stack

• Node.js and Express: Backend application and API routes.
• Vanilla JavaScript, HTML, and CSS: Frontend interface.
• Ollama and Gemma: Local inference for file-organization recommendations.
• SHA-256: Content-based duplicate detection.
• JSON Lines recovery journal: Persistent tracking of organization operations.
• Node.js test runner: Automated unit and end-to-end tests.
• GitHub Actions: Continuous integration.

The AI receives a limited catalog of file metadata rather than the contents of users' documents. Its recommendations are validated against known files, and it cannot directly execute a move simply because the model suggested one.

The organization workflow separates preview from execution. Approval tokens expire, paths are revalidated, destination conflicts are checked, and the application records operation states so interrupted work can be flagged for review.

The project also includes automated tests for scanning, duplicate detection, organization workflows, replay prevention, and recovery scenarios. The latest development report recorded 66 passing tests; this should be confirmed against the final repository before publication.

Why Does Open Innovation Matter?

Digital organization can involve deeply personal information: financial documents, photographs, work files, and private records.

I wanted filoAI to explore an approach where users can run AI locally instead of automatically sending their file metadata to a third-party AI service.

Using an open-weight model through Ollama gives the project flexibility. Users can choose a compatible model, inspect the application code, and adapt the system to their needs.

Local inference also provides a way to reduce dependence on remote AI APIs. When the model and runtime are already installed, recommendations can be generated locally without requiring a cloud AI provider.

Open innovation matters here because privacy, transparency, and user control should be part of the design—not features added after the product is built.

The Lazy Payoff

The most repetitive part of organizing files is not necessarily moving them. It's figuring out what is there, finding duplicates, deciding where things belong, and checking that nothing important gets overwritten.

filoAI brings those steps into one workflow.

Instead of inspecting every file manually, a user can scan an approved folder, review duplicate groups, receive suggestions, and approve selected moves.