Natural Language Local File Search

Describe a file in plain English on your Mac, and a local AI model finds it without uploading

Problem

Developers and other Mac users often forget file locations and resort to fuzzy memory while digging through Finder. Existing search tools either rely on exact filename matches or require cloud uploads, compromising privacy and speed. Someone tried with a local small model, but there's still room to improve accuracy and usability.

Solution

A Mac menu bar app where users type a natural language description (e.g., 'the spreadsheet about budget modified last week'), and a locally running lightweight model parses the intent and locates files. Results appear instantly, with options to open, preview, or copy the path. All processing stays local, no internet required.

Users

Mac users who frequently search for files, especially developers, designers, and knowledge workers.

MVP

Integrate a fine-tuned small language model (e.g., Qwen2.5 1.5B) to generate search plans; leverage macOS Spotlight index and file metadata for search execution; provide a simple natural language input box and result list; runs fully offline; supports common file types.

Revenue

One-time purchase (e.g., $15) or subscription (with updates and advanced models); distribute via Mac App Store or independently.

Why now

Local large model inference is now feasible with Apple Silicon; user demand for privacy and offline tools is rising; indie developers can leverage open-source models to build quickly.