# SlothSort AI: Let Gemma Clean Your Messy Files So You Don't Have To > Source: > Published: 2026-10-25 15:41:28+00:00 *This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady.* SlothSort AI is a local-first AI file organizer that turns the repetitive job of cleaning up a messy folder into a simple review-and-apply workflow. Instead of manually sorting files one by one, SlothSort lets local Gemma generate an organization plan that the user can review before anything changes. The workflow is: SlothSort does not read file contents. It uses metadata such as filenames, extensions, file sizes, and modification times to create the organization plan. [Watch the full SlothSort AI demo](https://drive.google.com/file/d/13X1HWn7zbp1en4gDqcEbAZfpJyRiqz6b/view?usp=sharing) The demo shows the complete workflow from creating a fake folder and scanning its files to generating, reviewing, and applying the AI organization plan. [View the SlothSort AI source code on GitHub](https://github.com/MANAV-MISHRA-BYTES/slothsort-ai) SlothSort AI is built with: The architecture is intentionally lightweight: ``` User ↓ SlothSort UI ↓ FastAPI ↓ Ollama ↓ Gemma 4 E4B ↓ Structured JSON plan ↓ Pydantic validation ↓ Review ↓ User confirmation ↓ Filesystem changes ``` The backend scans the selected folder and collects metadata from visible files. That metadata is sent to Gemma through Ollama running locally. Gemma returns a structured organization plan, which is validated with Pydantic before being presented to the user. Nothing is changed during scanning or planning. Files are only moved after the user reviews the proposed changes and explicitly confirms the operation. The backend also rechecks the scan before applying changes, keeps destinations inside the selected root folder, avoids intentional overwrites, and never deletes files. SlothSort uses Gemma locally through Ollama instead of depending on a proprietary hosted AI API. This means the application can perform its AI planning locally without requiring a cloud AI account or API key. File contents are never read, and only file metadata required for organization is provided to the local model. Open-weight models make this kind of local-first application possible while giving developers and users more control over the AI component and where their data is processed. Organizing a messy folder is exactly the kind of repetitive task that is easy to keep postponing. SlothSort turns: ``` Look at file → Decide → Rename → Move → Repeat ``` into: ``` Select folder → Scan → Review AI plan → Confirm ``` The goal is simple: let local AI handle the repetitive first pass while keeping the user in control of the final changes. AI coding tools were used during implementation, debugging, and refinement of the project. The project concept, product direction, feature decisions, testing, final review, and submission were human-led. Solo submission.