Basically, I didn't want to waste my opus 5.5 or codex Astra credits just to find files in my machine, so I trained a local LLM which is only 1.5B to write bash to find files on your macos laptop (works on apple silicon, uses little ram).
Full privacy since it's local so you can look for your sensitive files and api keys without sending them to a third party server.
It outperforms strong cloud models like Gemini 3.8 Flash and hefty 31B Param Gemma models in the file search benchmark I created to measure competency, details here: https://shlexball.com/findbench/
Comments URL: [https://news.ycombinator.com/item?id=49958869](https://news.ycombinator.com/item?id=49958869)
Points: 1
source & further reading
shlexball.com — original article