How I Had an LLM Organize My Obsidian Notes

Six months ago, it seemed like the problem had been solved once and for all: a well-thought-out folder hierarchy modeled after PARA, templates, and tags—the perfect system for storing knowledge. For the first few weeks, everything worked as intended.

But after a few months, the structure didn’t fall apart—it simply fell asleep. Notes kept appearing, but sorting them into the right places became, frankly, a chore.

Why the Sorting Ritual Is Exhausting

Every new note required a small set of decisions:

• where to put it;
• what to name it;
• what to link it to;
• which tags to assign.

Individually, any of these decisions takes a couple of seconds. But when there are dozens of such decisions to make every day, the brain starts to resist. As a result, the notes end up in one big pile, and the neatly organized hierarchy turns into a museum—beautiful, but uninhabited.

It’s Not Just Personal Laziness

The most unpleasant discovery was that this problem isn’t unique to me. According to the author’s observations, most people who try to maintain a knowledge base face this issue. Moreover, some users reject the idea of systematic note-taking altogether from the start, fearing precisely this outcome: “I’ll start—and in a month, I’ll end up with a dump of three hundred files with no logic whatsoever.”

The Idea: Let the Model Handle the Organization

This personal frustration with manual routines gave rise to an idea—to offload routine decisions about note categorization onto a language model. Instead of manually selecting a folder, name, and tags every time, this task can be entrusted to an LLM built directly into the Obsidian workflow.

This is how the idea for my own plugin was born—one that takes on the role of an “archivist,” analyzing the note’s content and helping to fit it into the existing structure without human intervention at every step.

Source: All Articles / Machine Learning / Habr