Shlexball

Local 1.5B model finds files by description on Mac, without privacy leaks.

Problem

Developers don't want to burn cloud LLM credits or leak sensitive file paths just to locate a few files on their Mac. A Hacker News post describes training a local 1.5B model specifically to convert natural language into bash find commands, proving the demand. But downloading models, configuring environments, and writing scripts is too high a barrier for most users.

Solution

An out-of-the-box desktop app with an optimized 1.5B model. Users type requests like 'all .env files modified last week' and it generates and runs safe find/mdfind commands. It shows the command and expected results before execution, with one-click open in Finder. Fully offline, also supports swapping in the user's own local model.

Users

Privacy-conscious developers, sysadmins, security engineers, and Mac users who frequently search for sensitive local files without wanting to send data to the cloud.

MVP

Install-and-run desktop app; built-in 1.5B model; natural language to find command; command preview and confirmation; result list with Finder jump; no network requests.

Revenue

One-time purchase $19.99, or $2.99/month subscription for ongoing model updates and advanced filters; team bulk license available.

Why now

Small local models run smoothly on Apple Silicon, while cloud AI quota and privacy costs are rising, letting indie developers quickly wrap this into an easy tool.