SlothSort: Channeling Our Inner Sloth with 100% Offline Open-Source AI

Krishna5567 — Sun, 25 Oct 2026 06:18:52 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

Sloths move at their own unhurried pace, conserve their energy, and never expend calories when they don't have to. Human developers, on the other hand, spend hours:

Drowning in chaotic Downloads and Desktop folders filled with cryptic files (scan_p10.png, step_576.txt, temp_scratch_v2.py).
Dreading listening to rambling 4-minute voice notes from colleagues.
Procrastinating on digital housekeeping because manual sorting and renaming is exhausting.
Hesitating to use cloud AI tools because they demand recurring subscriptions and upload private bank receipts, personal notes, and voice recordings to remote servers.

To celebrate International Sloth Day (October 20) and reward peak laziness, I built SlothSort—an autonomous local desktop and voice memo organizer powered 100% by open-weight models.

How It Rewards Peak Laziness:

Drop & Sleep: Dump your cluttered downloads, invoices, code snippets, and voice memos into a folder (or use the built-in Lazy Sandbox).
Hit "Hibernate & Sort (Lazy Mode)": The sleepy sloth agent takes over. It transcribes audio recordings using local OpenAI Whisper, parses documents and code using Hugging Face's open-weight SmolLM2-360M-Instruct, sorts files into categorized folders, and creates companion .txt transcripts for every audio file.
Wake Up to a Morning Digest: When you wake up, your clutter is organized, cryptic filenames are cleanly renamed in snake_case, and you get an executive Morning Digest with a spoken audio briefing.
Zero-Stress Undo: If you ever change your mind, a 1-click rollback engine restores every single file to its exact original location.

Demo

🌐 Live Cloud Interactive Showcase: https://slothsort.vercel.app
💻 Local Hardware Mode: Runs 100% offline with CUDA acceleration on an NVIDIA RTX 3050 Laptop GPU (or CPU) taking under 1.8 GB VRAM.

Quick Tour of the Web UI:

🗂️ File Organizer: Live before/after inspection table showing smart renaming, categorization pills (Receipts & Finance, Voice Memos, Code & Scripts), and AI summaries.
☕ Morning Digest: Personalized sloth greeting, task statistics, energy conserved percentage, and audio playback with browser speech synthesis.
🎙️ Voice Memos & Whisper: Side-by-side audio playback and synchronized Whisper transcripts.
↩️ Rollback History: Transaction-level rollback manifests with 1-click undo.
🌱 Open Innovation Manifesto: Real-time hardware specs, zero cloud cost badges, and offline architecture proof.

Code

Krishna5567 / SlothSort

Slow and Steady: Autonomous 100% offline local AI desktop & voice memo organizer (SmolLM2 + Whisper + pyttsx3)

🦥 SlothSort: Slow & Steady Autonomous Local AI Desktop Tidier

Built for the "Slow and Steady" Hackathon Challenge (International Sloth Day Edition)
Channeling our inner sloth to automate digital clutter with 100% open-source AI, zero cloud dependencies, and zero cost.

🍃 Overview

Sloths don't stress, they don't rush, and they conserve their energy. Human developers, on the other hand, spend hours:

Digging through chaotic Downloads/ and Desktop/ folders.
Listening to rambling 5-minute voice memos from colleagues.
Renaming cryptic files like scan_p10.png or IMG_2984.PNG.
Dealing with paid cloud AI subscriptions that fail as soon as the Wi-Fi drops.

SlothSort is an autonomous desktop and voice note organizer powered entirely by open-weight and open-source AI models running locally on your machine.

You dump your cluttered files into a folder, hit "Hibernate & Sort (Lazy Mode)", and go to sleep. SlothSort steadily transcribes audio recordings using local OpenAI Whisper…

View on GitHub

👉 GitHub Repository: https://github.com/Krishna5567/SlothSort

How I Built It

SlothSort is designed as a modular, local-first agent harness:

openai-whisper (base model, MIT License): Performs on-device automatic speech recognition. Transcribes voice notes directly on GPU in seconds without sending a single byte to the cloud.
HuggingFaceTB/SmolLM2-360M-Instruct (Apache 2.0 License): A lightweight 360M parameter open-weight model that runs in FP16 on consumer laptop GPUs (under 1 GB VRAM). Handles semantic document categorization, smart snake_case renaming, and generates the Morning Digest.
pyttsx3 & Web Speech API: Generates native offline audio briefings (morning_digest.wav) with zero external API calls.
FastAPI & Uvicorn: High-performance asynchronous backend with structured Pydantic schemas.
Vercel Edge & Serverless Architecture: Optimized with a lightweight serverless profile (< 20 MB bundle size) and inlined glassmorphic styling for instant web deployment.

Why Does Open Innovation Matter?

1. Absolute Data Sovereignty (Keeping Personal Files Off Remote Servers)

A desktop organizer handles the most intimate data in your digital life: bank statements, tax invoices, medical notes, private voice memos, and proprietary code. Sending these files to a closed cloud API creates severe privacy and security risks. With local open-weight models, 0 bytes of user data ever leave your machine.

2. 100% Offline Capability (Airplane & Cabin Friendly)

Closed APIs completely brick the moment your Wi-Fi drops. SlothSort runs on a laptop GPU (taking under 1.8 GB VRAM) or on standard CPU without needing any internet connection.

3. $0.00 Marginal Cost Forever

Closed LLM and Speech APIs charge per input token, output token, and audio minute. SlothSort costs $0.00 forever with zero subscriptions and zero API keys.

4. Zero Vendor Lock-in & Model Freedom

Because SlothSort is built on open standards, you can swap Whisper models, quantize SmolLM2 with GGUF, or adapt prompts freely without being bound to vendor deprecation cycles.

My Agent Session

This project was architected, implemented, and deployed using autonomous agent pair-programming:

Complete Project Workspace & Code: https://github.com/Krishna5567/SlothSort
Local Hardware Verified: NVIDIA GeForce RTX 3050 Laptop GPU (CUDA 12.4 + PyTorch 2.6.0)

The Lazy Payoff

Real-World Testing & Time Saved:

During testing this week, I dumped a cluttered directory containing messy receipts (scan_p10.txt), raw group chat exports, python benchmark scripts, and voice memos:

Manual Sorting Time Avoided: ~35 to 45 minutes of dragging, reading, listening, and renaming files.
Audio Friction Eliminated: Rather than sitting through audio recordings, SlothSort generated a clean companion .txt transcript right beside each audio file in seconds.
Effort Expended: Exactly 1 click on "Hibernate & Sort (Lazy Mode)".
Human Energy Conserved: 100%.

Prize Categories

Primary Challenge: Open-Source AI Challenge Week 3: Slow and Steady (International Sloth Day Edition)
Sub-themes addressed: Rewarding laziness, running open-weight models locally, keeping personal data off cloud servers, and automating everyday clutter with zero running cost.

DEV Community: Krishna5567