Arrange My Files AI: A Clearer Home for Every File

A smarter, safer way to organize your files — without losing control of your documents.

Arrange My Files AI turns messy folders into clear, reviewable organization plans, with verified copies and original files preserved.

A folder full of study notes often has two problems. Some filenames tell us almost nothing: 0.docx, final2.pdf, or notes.txt. Others are already useful, but scattered across unrelated folders. Renaming everything would be unnecessary; ignoring the difficult files would leave the job unfinished.

Arrange My Files AI creates an organization plan, lets you review it, and copies the included batch into a new output session. Clear plans go into category folders. Uncertain or unsupported files go unchanged into Needs Review. Originals remain in place.

The intended user is someone with accumulated course materials, project documents, or office notes who wants a traceable result without opening every document first.

![The deployed homepage introduces the private document organizer.]

Fresh live-site screenshot, captured on a 1440 × 1000 desktop viewport.

Try the deployed application: Arrange My Files AI.

Start with How it works for a simulation, or open Organize Files with a small disposable test batch. Synthetic examples inside the app are previews; they do not perform real output copying.

You can drop multiple files or folders together, select several files through Add files, or use Add folder repeatedly. New inputs accumulate in the current batch.

For this walkthrough I generated nine dummy files: DOCX, PDF, PPTX, XLSX, CSV, Markdown, a plain-text checklist, an empty text file, and a small PNG. None contains personal information.

![The workspace explains adding documents and the batch limits.]

This live run used Add files. It did not test recursive folder dropping. The current limits are 1,000 files in a combined batch, 12 nested folder levels, and 20 MB per file for content analysis. Larger readable files can still be copied unchanged into Needs Review. An incomplete scan is explicitly reported rather than presented as the whole input.

Choose Build organization plan. The engine looks for useful filenames, document title metadata, headings and a bounded opening excerpt. Editable keyword rules choose category folders.

The live nine-file run produced five category plans and four Needs Review files. All nine stayed included. The UI displayed zero elapsed seconds for this small run; that is rounded UI output, not a precise performance benchmark.

![The live organization plan accounts for the nine-file batch.]

The naming behavior is selective. 0.docx became Project Delivery Roadmap.docx, based on its title. final2.pdf became Study Skills Workbook.pdf. A descriptive name such as Project Launch Checklist.txt was preserved.

![A recovered document title explains the proposed DOCX name.]

This is quick organization, not full-document understanding. PDFs use their first two pages; Word uses opening content. Slides and workbooks also use bounded samples. The rules inspect at most 3,600 characters for naming evidence.

An empty file cannot supply a useful title. A PNG has no supported image analysis in this release. Some otherwise readable files may still require review. These cases do not disappear from the batch: their original names are retained under Needs Review.

![The Needs Review filter shows uncertain files while keeping them included.]

The count buttons filter the list. They do not change which files will be copied. Unchecking a file deliberately excludes it. Accepting a category plan is a separate choice from including that file.

![An earlier local browser test demonstrates category and review routing.]

Archive evidence from an earlier local test, not the fresh nine-file live run.

In the normal workflow, Copy all N files asks for a separate output folder and confirms the batch. A new session folder contains category copies, Needs Review copies and an operation manifest. Each successful copy is reopened and compared with the original SHA-256 hash.

An earlier local browser acceptance run processed 131 synthetic files: 83 organized + 48 Needs Review, with 131 verified copies and zero failures. It also counted all 131 source files afterward. That was controlled local testing, not a claim about every laptop or real-world collection.

![Earlier local acceptance evidence shows 131 verified copies.]

The checker maps sources to destinations. Recheck saved copies reads output bytes again to detect missing or changed copies. It does not decide whether a category was appropriate. The CSV report provides a spreadsheet-friendly audit trail; reopening the saved output folder enables checking after a refresh.

![An earlier copy-checker test shows source-to-destination tracing.]

Incomplete work must also be visible. Failed and pending/not-recorded counts matter, even when existing copies verify correctly. Automatic resume is not available in this release.

![An earlier interrupted-copy report makes pending work explicit.]

Fresh-test limitation: the live planning and naming steps above passed. The automated browser could not complete the Windows output-folder selection, and the desktop fallback stopped because it could not reliably verify the active browser URL. This walkthrough therefore does not claim a new successful live copy/recheck run. The copy screenshots are explicitly labeled earlier evidence.

Repository: Navid-Ishaq/arrangemyfilesai-detleng.

The application uses React and TypeScript, browser document readers, a worker-based planning path, editable literal keyword rules, and the File System Access API for output operations. It handles document reading separately from copy verification.

The biggest design change was removing model inference from the critical organization path. Earlier experiments made ordinary document batches too slow to feel useful. The current engine performs bounded document reading and deterministic naming/category decisions locally.

The DeTLeng AI Guide has a different job: explaining the workflow and known warnings. Its backend uses the open-source Mastra agent framework, a curated help tool, and Groq inference. The configured default model in the source is openai/gpt-oss-20b; deployment configuration can override it. Render hosts the Guide API. The Guide does not inspect, rename or copy the selected documents.

This separation is intentional: file operations should not wait for a conversation. The online Guide receives chat messages and optional stage/error codes, not the private file inventory automatically.

Readable source code and inspectable rules make the organizer's decisions easier to explain and improve. Mastra exposes the Guide's instructions and tool boundaries in code, while the model configuration can be changed independently of the document-copy engine.

There is also an important challenge-fit distinction: this project's organizer is deterministic, and its open-source AI component is a separate help layer. I would not describe it as AI-driven file organization. Whether that satisfies an “open-source AI at its core” competition requirement needs organizer confirmation; adding an optional chat interface alone should not be presented as proof of eligibility.

The useful outcome is less repetitive checking: one batch, proposed destinations, a review holding area, and a traceable report. There is still a human decision where evidence is weak. The promise is a clearer workload, not magical understanding of every document.

![The interactive walkthrough explains the complete batch before users try their own files.]

Earlier desktop capture of the built-in simulation. It performs no real file copying.

Before treating this as competition-ready, I want a fresh manual desktop output-copy/recheck run using the included dummy batch, confirmation of challenge eligibility, and verification of repository licensing and build-window dates. Those are concrete release checks, not reasons to exaggerate the current evidence.

Built by Muhammad Naveed Ishaque at DeTLeng, with AI-assisted development using Codex. My aim is simple: make everyday file organization clearer, safer, and less tiring.

The core file-organization workflow uses transparent, deterministic rules for speed and predictability. The separate DeTLeng AI Guide, built with the open-source Mastra framework, provides conversational guidance and troubleshooting.

The Guide does not access, analyze, rename, or modify private files. Document organization remains a separate, locally executed workflow, with users retaining control over their files.