NeatAI / FileMind
An AI-Powered Second Brain For Your Computer
Auto
Categorization
0
Cloud Uploads (Local)
Smart
Duplicate Scan
Executive Overview
NeatAI reimagines file organization as an intelligent, on-device experience. Built on a Tauri 2 + Rust core with a React 19 / TanStack Start frontend, the system pairs a high-performance multithreaded filesystem engine (Rayon + SHA-256) with a fully local AI pipeline — Ollama-served LLMs for classification, Tesseract.js for OCR, and a custom Float32Array vector database for semantic search. Every destructive operation is reversible via a SQLite undo journal.
The Problem
Users waste hours searching for files, dealing with duplicates, and manually organizing folders.
The Solution
Semantic file organization using local LLMs to categorize documents, detect exact/similar duplicates, and enable natural language file search.
Core Capabilities
Semantic Organization
Auto-sorts files into context-aware smart folders.
Duplicate Detection
Finds identical files and visually similar screenshots.
Natural Language Search
Chat with your files using local vector embeddings.
OCR & Document AI
Understands text inside images, scanned PDFs, and receipts.
Interface Gallery
Dashboard
Technical Deep Dive
Semantic Categorization Pipeline
Files are processed through a multi-stage pipeline: Extension → MIME type → Content extraction (OCR for images, text for docs) → LLM classification → Smart folder assignment with confidence scoring.
Vector Search Architecture
Each file generates a 384-dim embedding via all-MiniLM-L6-v2. Stored in ChromaDB with metadata filters. Natural language queries are embedded and matched via cosine similarity with re-ranking.
Safe Cleanup Engine
Deletion candidates are staged in a virtual recycle bin with full undo capability. Each cleanup operation is logged as a reversible transaction with file checksums for integrity verification.