New Plugin: Beets AI Sauce

[semohr]

Beets AI Sauce

We gave AI your metadata. It panicked, guessed, and somehow made it better. Beets AI Sauce uses AI to identify, clean, and enhance your music metadata.

Key features:

• AI-powered metadata suggestions & corrections - let the bots do the boring bits
• Automatic clutter cleanup - strip out "FREE DOWNLOAD" tags, ALL CAPS madness, and other noise
• Fully customizable prompts - tailor the AI’s behavior to your collection’s quirks

What started as a joke between @pSpitzner and me turned out to be surprisingly useful, especially for cleaning up dubplates, bootlegs, or adding metadata to your own tracks.

GitHub: metasauce/beets-aisauce
PyPi: beets-aisauce

Note: Currently works only with the latest beets master branch (>2.3.1) due to the new Metadata plugin architecture, but fingers crossed for a release soon! This is still a fresh experiment and only supports rudimentary features for now. Expect some rough edges, occasional AI weirdness.

[arsaboo]

This looks great @semohr

I have a version of this in my Plexsync plugin, where I combine search (currently Tavily, Exa, and SearXNG are supported) with LLM-based parsing (using Ollama). See llm.py for implementation details.

[halloleo]

Wow, cool. I will check it out!

I would like to try it with my local ollama. Should that work?

[semohr]

Yeah, olama has an OpenAI compatible API endpoint so should work without issues. Depending on the model the results may vary tho.

[semohr]

Quick update: the plugin can now also be used to clean up metadata before the import process. This means it can be used similarly to the fromfilename plugin to fix broken metadata ahead of the actual autotagging step.