unary-works/unfoldly

Unfoldly is a local desktop application that lets you search your personal files using natural language, the way you actually remember things, rather than requiring exact filenames or keywords. Instead of asking what a file is called, you can describe what you remember about it: the photo of your dog on the beach, the video where you saw a manta ray, or the spreadsheet that explains why a trip went over budget. The app works by indexing the files and folders you choose, extracting text, visual signals, metadata, and searchable embeddings from each file type. Once indexed, you can search by memory or meaning, and then ask follow-up questions about the file you find, with the app keeping the source document in context for the conversation. The full search and chat process runs locally, so your files are never uploaded anywhere. Supported file types span a wide range: documents such as PDF, DOCX, and TXT, spreadsheets including XLSX and CSV, slides in formats like PPTX and Keynote, images in JPG, PNG, HEIC, and others, audio files, video in formats like MP4 and MOV, and structured files like JSON and YAML. Under the hood, Unfoldly runs a Python backend with ChromaDB as its local vector store, uses local GGUF format AI models for language understanding, and wraps everything in a Tauri based desktop shell with a frontend built in TypeScript. The current public release is for macOS only and is in early beta. The full README is longer than what was shown.