Local AI Toolkit

Local AI is powerful, and rough exactly where daily life happens

Running models locally is great — until your machine grinds to a halt because a 14B model you forgot about is still loaded, or until your screenshots folder holds 800 files named Screenshot 2026-03-14 at 09.43.11.png and nothing in them is findable.

These are not model problems; they're living-with-models problems. Nobody was building for them, so I did — three small macOS utilities, each doing exactly one job, all running entirely on-device. No uploads, no cloud, no accounts.

Three tools, one job each

• mini-tagger

  A menu-bar watcher: every new screenshot goes to a compact vision model running on the Apple Silicon GPU, comes back with semantic tags, and lands renamed — “(terminal python error traceback)” — in the right folder, name already on your clipboard.

• mini-renamer

  Human-in-the-loop renaming: drag in images, PDFs, or DOCX and a local LLM proposes a clean filename for each. You review, edit inline, apply. Naming style — kebab-case, snake_case, Title Case — is yours to configure.

• lms-guard

  A memory watchdog for LM Studio that unloads idle models before your system starts swapping. The most boring tool of the three, and the one that saves the day most often.

The craft is in the edges. macOS writes screenshots via atomic temp-file renames, so the watcher handles both created and moved events; a stability loop waits for the file size to settle before the model ever sees a partial write; and mini-tagger monitors its own memory footprint so the tool that manages your RAM never becomes the problem it solves.

Infrastructure you stop noticing

All three tools run daily on my machine, quietly. Screenshots arrive pre-named and searchable, downloads get sensible names in seconds, and the model that would have eaten the afternoon's RAM was unloaded before I noticed it was idle.

The best measure of a utility is how rarely you think about it. These three earned their silence — and my files finally have names that mean something.