NEVER SORTED
Inspiration
I have a 12TB NAS at home, and it is a digital dumping ground. I’m a digital hoarder—and if we are being honest, most NAS owners are too. We back up years of mismatched folders, downloads, blurry phone bursts, 2017 tax documents, and duplicate receipts with the promise of "sorting it next weekend." But next weekend never comes because organizing files is tedious, and uploading multiple gigabytes of incredibly personal data to a cloud-based AI to classify it feels like a massive privacy violation.
The inspiration for NEVER SORTED was born out of frustration and an essential truth: I want my files organized, but I want my private data to stay entirely on my hardware. With UGREEN opening up an AI NAS, we realized we could build an intelligent, background digital archivist that acts like a human organizer—doing the heavy lifting on-device without ever talking to the cloud.
What it does
NEVER SORTED is a privacy-first, autonomous file organizer designed entirely for local UGREEN AI NAS hardware. It works like a diligent digital housekeeper:
- Intelligent File Sorting: It monitors selected directories and automatically runs local OCR and image-analysis models to determine what a file actually is. No more guessing by file names like
IMG_4829_edited.jpgordraft_v3.pdf. - Context-Aware Renaming: It analyzes content (such as a date and vendor on a receipt, or the exact year of a tax form) and automatically renames the files logically (e.g.,
2026-11_Uber_Receipt_$15.42.pdf) and moves them to secure, dedicated folders. - Local Media Deduplication: It analyzes photos and media to differentiate actual memory keepers from clutter like screenshots, memes, and burst shots—quietly highlighting high-quality photographs while group-clustering duplicates so users can easily reclaim lost space.
Everything is processed completely locally. It is the first hands-free organizational assistant that guarantees you never have to sort a file again—while retaining absolute data sovereignty.
How we built it
We had to keep it lightweight, fast, and secure. We utilized the UGREEN Skill Developer Package as our foundation to integrate directly with UGOS.
- The Core: We built a local orchestrator that triggers whenever new files land on the NAS.
- The Brains: We integrated tiny, optimized on-device models: lightweight Vision Transformers for processing images/memes, and efficient local OCR for extracting text from PDFs and receipts.
- Smart Pipelines: The system relies on local SQLite indexing to map, verify, and trace files without overloading CPU memory. We developed "Skill" automation scripts using Python, working in tandem with the UGOS-supported command-line tools to move, rename, and tag assets without creating processing bottlenecks.
Challenges we ran into
Building AI to run on local, consumer-level hardware is a totally different ballgame than calling a giant cloud API.
- Resource Constraints: NAS devices aren't packed with massive server-grade GPUs. We ran into heavy CPU throttling during our early tests when trying to parse hundreds of photos at once. We had to drastically optimize our workflows—using chunked queues, batching file reads, and scheduling intensive image processing to run when the NAS is idle (like 3 AM) to keep the system responsive.
- File Clashes & Edge Cases: How do you handle file conflicts safely without deleting anything? Building safe, non-destructive file movements, backup fallbacks, and handling weird nested directories where our file watcher got stuck in infinite loops was a massive headache.
- Making No-NAS testing works: Building the entire environment locally without UGREEN hardware in Round 1 was tricky. We spent a lot of time simulating hardware environments and ensuring our container boundaries didn't depend on proprietary drivers.
Accomplishments that we're proud of
- Zero Cloud Dependency: We successfully proved that you don't need to pay OpenAI or worry about data leaks to have smart, context-rich automation. It’s entirely local, runs on the device, and respects user privacy.
- Intuitive Simplicity: We made it feel like magic. You literally drag a messy folder of 500 unsorted files in, and within minutes, they are neatly categorized into chronological folders, financials, and OCR'd receipts with sensible names.
- The Cute Persona: We designed a personality around it. It feels less like an intimidating system script and more like a helpful little utility (hence our cute mascot!).
What we learned
- Local is Hard, but Worth It: Optimize, optimize, optimize. We had to rely on small, efficient open-source models (and prune them where we could) to make them fast enough on small CPU clusters. Less is more.
- The Value of Edge AI: Keeping data local changes your perspective on building software. Standard dev practices rely so heavily on external APIs; going "off-grid" forced us to write much more robust, self-contained, and failure-tolerant code.
What's next for NEVER SORTED
If selected for Round 2, our absolute priority is deploying NEVER SORTED onto a live UGREEN AI NAS iDX6011 Pro.
Specifically, we want to:
- Hardware Polish: Maximize hardware-accelerated processing by fine-tuning the resource footprint directly on the UGOS kernel.
- True Collaborative Workflows: Let users review changes via a "Weekly Digest" skill card inside the UGOS interface, offering a simple click-to-approve interface for deduplication suggestions.
- Continuous Local Learning: Implement tiny feedback loops where the AI can adapt to how a specific user likes their personal tagging style, over time creating a highly customized organizational assistant unique to every household or business.
