Understanding Obsidian in 24 Minutes: From Local Folder to AI-Powered Knowledge Universe

Obsidian began in 2020 as a minimalist note tool, yet its natural fit for AI has pushed it toward an estimated 5 to 10 million users in recent years; Tina Huang promises to compress hours of trial into 24 minutes, first racing through the app's skeleton and then demonstrating three tiers of AI use with live runs, ending with a short recall quiz to lock in the learning.

What Obsidian is: the single-folder philosophy

Everything starts with one local folder: open it in Obsidian and you get a vault, a directory whose contents are plain .md markdown files that stay human-readable and portable across Dropbox or Drive; Obsidian Sync is the one major paid add-on, end-to-end encrypted, keeping that same folder in sync across devices with version history included.

Pricing is straightforward: the app itself is free and unlimited, Sync runs about 4 dollars per month when billed yearly or 5 dollars monthly, with encryption and history bundled; because files live locally, ownership stays with you, which is why even the company notes it does not track total downloads and why privacy feels structural rather than marketed.

Links, graph and plugin universe

What made Obsidian famous was the double-bracket link: mentioning another note creates it, unwritten stubs appear in light gray, the local graph shows neighbours of a note while the full graph animates the entire vault, and filters let you group that network by theme, turning a folder into a navigable map of ideas.

Powerful search by path, tag and content, a vast community plugin library including Smart Connections and Copilot, aesthetic themes, and the ability to embed almost anything — images, PDFs, audio, tables, tweets, code blocks — turn the vault into a custom workbench; flexible metadata can morph into tables, Kanban boards or dashboards, and the command palette keeps navigation fast with hotkeys.

Why a non-note-taker chose Obsidian

Tina offers a candid confession here: she rates herself a 3 out of 10 as a note-taker, prefers dumping learning into a single Google Doc, and did not adopt Obsidian for daily journaling; what pulled her in was how effortlessly AI can read the vault, and she invites viewers to spot the three compatibility clues hidden in the feature tour.

Those three clues surface between the lines: plain markdown files, the folder-as-database structure, and an open plugin architecture; together they let an AI agent parse the vault without proprietary APIs, which becomes the foundation for the three levels that follow.

Level 1 — AI Second Brain

Level one is the simplest: you keep writing the notes, then point an agent such as Claude Cowork or Codex at the vault; Tina queries her own second brain about video concepts, the agent reports it read the whole vault, cites the file that lists six filtering criteria, and even builds a checkable checklist to decide what is worth filming.

The idea builds on Tiago Forte's second brain thesis that the mind is for thinking, not storing, so you write that Bobby's favorite color is yellow and retrieve it later; in the classic version you search yourself, in the AI version the model retrieves, interprets and advises, turning scattered notes into a conversational memory you can interrogate.

The tool palette is broad: Claude Cowork for a gentle UI, Codex and Claude Code for programmatic building of apps and dashboards, fully private harnesses such as Hermes or DeepSeek with local models, plus in-vault plugins that query directly; Tina switches deliberately between them depending on whether the task is asking or building.

Level 2 — AI Database and Level 3 — LLM Wiki

Level two flips the roles and lets AI do the writing: Life Bot logs an unsweetened green tea as zero calories from a photo, Taco Bot drafts process docs for a YouTube topic engine, desktop widgets push Pomodoro logs and to-dos automatically, so a rich personal and company archive accumulates without manual note-taking and the same agents can then read it back to advise on deep-work sequencing or break timing.

Under the hood the vault stays synchronized across a laptop, a Mac Studio and a VPS via Obsidian Sync, while writing is handled by Gemini Flash, DeepSeek V3 and, for fully local privacy, Qwen 36B on the Mac Studio; health data from Apple Health and a newly tried Oura ring flow into the same directory, giving both a human-readable archive and a queryable database.

Level three is the Obsidian interpretation of Andrej Karpathy's LLM Wiki pattern: treat the vault as an IDE, the model as the programmer and the vault as the codebase, with the model ingesting sources, managing pages and answering queries while you mostly observe; Tina's Hermes-based wiki ingests a Hermes Kanban guide, answers how to build a board by consulting its index and log, and periodically runs a health check to prune stale or contradictory pages.