Cat Finder: Personal Software via Vibe Coding

Personal Software via Vibe Coding: I needed to find cat photos scattered across my system. Instead of cloud services or app stores, I described what I wanted to Claude Code and got a working Rust CLI tool using YOLOv8 and ONNX Runtime. Privacy-first, locally-run, and mine to modify.

I needed to find cat photos scattered across my system. Instead of searching the app store, signing up for a cloud service, or uploading my personal photos to someone else’s servers, I asked Claude Code to build me the tool I needed. An hour later, I had it.

This is Personal Software—software that exists because you needed it, built the way you want it, running entirely under your control.

Resource	Link
Repo	cat-finder
Video	Cat Finder Explainer

The Vibe Coding Approach

Vibe Coding is about describing what you want and letting AI handle the implementation details. No boilerplate, no Stack Overflow rabbit holes, no fighting with build systems. You focus on the what, the AI handles the how.

For Cat Finder, the conversation went something like:

“I want a CLI tool that scans directories for images containing cats. Run locally, no cloud. Use YOLO for detection. Output just the file paths so I can pipe them to other commands.”

Claude Code chose the tech stack (Rust, YOLOv8n, ONNX Runtime), handled the tensor math, figured out the COCO class IDs, and produced a working tool. I guided the direction; the AI wrote the code.

Why Personal Software?

The traditional options for “find cat photos” would be:

Cloud service: Upload photos to Google/Apple/Amazon, let them scan everything, hope they respect your privacy
Desktop app: Find something in an app store, hope it does what you want, deal with subscription fees or ads
Write it yourself: Spend days learning YOLO integration, tensor formats, image preprocessing

Personal Software offers a fourth path: describe what you need, get exactly that, own the result completely.

Cat Finder runs entirely on my machine. No accounts, no uploads, no subscriptions, no ads. The code is mine to modify, extend, or share.

What Got Built

A Rust CLI tool using YOLOv8n (the nano variant) through ONNX Runtime:

Directory Traversal → Image Preprocessing → YOLO Inference → Cat Detection → Output

The Detection Pipeline

Walk directories recursively, finding image files (jpg, png, gif, webp, etc.)
Preprocess each image: resize to 640×640, normalize to 0.0-1.0, convert to NCHW tensor format
Run inference through the YOLOv8n ONNX model
Parse output for class ID 15 (cat in COCO ordering) above confidence threshold
Print matching paths to stdout for easy piping to other tools

Unix Philosophy

# stdout: just paths (machine-parseable)
# stderr: logging and progress

cat-finder ~/Photos | xargs -I {} cp {} ~/CatPhotos/

This separation enables composable Unix pipelines. The tool does one thing well and plays nicely with others.

Technology Stack

Component	Purpose
Rust	Memory-safe, high-performance core
YOLOv8n	Lightweight object detection (12MB model)
ONNX Runtime	Cross-platform inference engine
clap	CLI argument parsing
ndarray	Tensor operations
walkdir	Recursive directory traversal

Total footprint: ~80MB (runtime + model + binary)

I didn’t choose this stack—Claude Code did, based on the requirements. It made good choices.

Usage

# Basic usage
cat-finder ~/Photos

# Adjust confidence threshold
cat-finder --confidence 0.5 ~/Photos

# Verbose output with timestamps
cat-finder -v -t ~/Photos

# Copy all cat photos to a new folder
cat-finder ~/Photos | xargs -I {} cp {} ~/CatAlbum/

Honest About Limitations

The README documents failure cases transparently:

Image Type	Detection Success
Clear photographs	High
Artistic/stylized images	Low
Cats in clothing	Low
Small/partial cats	Variable
Low quality/blurry	Variable

Test results: 7 of 9 cat images detected (77.8% recall). Oil paintings and anthropomorphized cats confuse models trained on photographs. This is documented, not hidden.

Bonus Features

The project grew organically based on related needs:

Duplicate Finder: A second binary for finding duplicate images using size-based filtering followed by SHA-256 checksums.

find-duplicates ~/Photos

Web Demo: A Flask-based interface for visual feedback with real-time progress via Server-Sent Events.

These emerged from “while you’re at it…” requests during development. Vibe coding makes feature additions nearly frictionless.

Setup

git clone https://github.com/sw-ml-study/cat-finder
cd cat-finder
./scripts/setup.sh  # Downloads model, builds project
./cat-finder ~/Photos

The Personal Software Philosophy

Privacy-first: All processing happens locally. No cloud APIs, no external services, no data leaving your machine.

Ownership: The code is yours. Modify it, extend it, share it, delete it.

Fit-for-purpose: Built for exactly what you need, nothing more, nothing less.

Transparency: Known limitations documented. No marketing spin.

References

Resource	Link
YOLOv8	Ultralytics YOLOv8 - State-of-the-art object detection
ONNX Runtime	ONNX Runtime - Cross-platform inference engine
ort crate	ort - Rust bindings for ONNX Runtime
COCO Dataset	COCO Classes - Class ID 15 = cat

You don’t always need an app store or a cloud service. Sometimes you just need to describe what you want and let an AI build it for you. That’s vibe coding. That’s personal software.