Video summary

Obsidian in 24 Minutes

Main summary

Key takeaways

Product Review

Product reviewed: Obsidian (note-taking app)

The video positions Obsidian as a note-taking “vault” system that becomes especially powerful when paired with AI—useful for second brains, AI databases, and LLM wikis.

Key features highlighted (core Obsidian)

  • Vault-based storage: Everything lives in a single local folder (“vault”).
  • Markdown (.md) notes: Notes are easy to read and compatible with standard tooling since they’re normal .md files.
  • Local + syncing options:
    • Can sync with third parties like Dropbox / Google Drive.
    • Includes Obsidian Sync (called out as one of the only paid features), with end-to-end encryption and described as very secure.
  • Linking and knowledge graph:
    • Link notes using bracket syntax (e.g., [[Lonely Octopus]]).
    • Includes local graph view and full graph view (including aesthetic visuals).
    • Supports stubs (linked but not yet created notes).
  • Powerful search: Search by path, file, tags, and other criteria.
  • Embeds: “Embed pretty much anything,” including images, PDFs, audio, tables, tweets, code blocks, etc.
  • Metadata + transformations: Add multiple metadata types; transform notes into tables, Kanban boards, dashboards, etc.
  • Hotkeys/command palette: Strong emphasis on keyboard-driven navigation.

AI use-cases (3 “levels” of integration)

Level 1: AI Second Brain (AI retrieves + analyzes your existing notes)

  • You write notes normally into your Obsidian vault.
  • An AI is given access to the vault and can:
    • Answer questions after “reading the whole vault”
    • Do conversations with your notes

Examples

  • Using Claude / Cohere-type agents for Q&A over the vault.
  • Using Codex-type agents for more programmatic outputs, like generating a checklist from your notes.

Verdict for this level: Focused on retrieval/analysis over already-written information.

Pros mentioned

  • Turns markdown notes into an AI-readable knowledge source
  • Enables Q&A and deeper interpretation of your saved notes

Cons implied

  • You still need to populate and maintain your notes yourself

Level 2: AI Database (AI writes into your vault)

  • Reverses Level 1:
    • AI agents write logs and documentation into the vault
    • You query it for insights and advice

Examples in the video

  • “Life Bot” logs health/food inputs (example: green tea logging via image).
  • “Taco Bot” creates documentation (e.g., process docs for YouTube topic engine).
  • Desktop widgets write Pomodoro logs and to-do lists into the vault.

Benefits

  • AI can give specific productivity advice based on historical patterns (e.g., deep-work timing, task sequencing, break timing).
  • Can enable more productive teams by centralizing work/productivity data.

Multi-machine syncing

  • Uses Obsidian Sync so multiple machines/agents can write to the same vault.

Models mentioned

  • Gemini Flash
  • DeepSeek V4 Pro
  • Optional local privacy model: Qwen 3.6 36B (running locally)

Health example (side context)

  • Aura ring (compared against Apple Watch / Whoop as considered options).

Pros mentioned

  • Less manual note-taking; AI populates the database
  • Personalization based on real activity/health/work data
  • Works across multiple devices via syncing

Cons implied

  • Requires multiple AI agents/tooling and ongoing setup
  • You must manage model/tool choices (cloud vs local)

Recommendation callout

  • “Would recommend trying out this level first” for a rich personal data/notes repository.

Level 3: LLM Wiki (fully AI-managed knowledge base; minimal user touch)

  • Described as an architecture popularized/outlined by Andrej Karpathy.
  • Conceptual framing:
    • Obsidian vault = code base
    • LLM = programmer
    • You act as an observer

How the system works

  • AI ingests sources
  • AI writes/updates wiki pages
  • AI queries/synthesizes answers
  • AI periodically lint-checks (finds stale claims, contradictions, orphan pages, etc.)

Example workflow shown

  • A Hermes-related wiki built by a “Wiki Bot”
  • The wiki can answer “How do I…” questions step-by-step using stored pages
  • Includes a periodic “health check”/linting pass

The “3-layer” design (very explicitly described)

  1. Raw sources (immutable; AI reads but doesn’t modify)
  2. Wiki (LLM-generated markdown pages; AI owns this layer)
  3. Schema/config files describing structure (e.g., how the wiki is organized)

Special wiki files

  • index.md: catalog of pages + metadata
  • log.md: chronological append-only record of ingest/query/lint activity

Implementation practicality

  • Claimed can be set up in < 5 minutes
  • Uses Hermes (including a “LLM wiki” skill + Telegram bot as an ingest method)

Notes alternatives

  • Best-documented alternative: Claude Code (clonable GitHub templates)

Positioning/comparison

  • Claims Obsidian is the go-to for LLM wikis, and even Karpathy uses Obsidian.

Pros mentioned

  • Near “hands-off” knowledge organization
  • AI-maintained structure + automatic QA via linting
  • Designed to scale and stay coherent over time

Cons implied

  • More complex conceptually and operationally than Levels 1–2 (even if “easy to implement” today)
  • Still depends on external AI tooling (though it can be local)

Comparisons made

  • Contrasts manual/traditional second brain (human stores + human retrieves) with AI second brain (AI retrieves + interprets).
  • For LLM wiki tooling:
    • Mentions Hermes as an example implementation
    • Says Claude Code is the most documented alternative, with GitHub setups
  • For health tracking (side topic):
    • Compares Aura ring vs Apple Watch vs Whoop (video says it chose Aura ring)

(No direct head-to-head comparison against other note apps besides describing Obsidian as the go-to for this AI workflow.)

User experience (as described in the video)

  • The creator says they rarely take notes personally and typically use a Google Doc, but still chooses Obsidian due to AI compatibility.
  • Obsidian is portrayed as:
    • Readable (markdown)
    • Navigable (graphs + search + embeds)
    • Highly extensible (plugins/themes)
  • AI integration experience:
    • AI can “read the whole vault,” answer questions, and generate checklists/docs
    • In Level 2/3, AI writes into the vault via tools/agents, making retrieval more useful than manual capture

Pros and cons (consolidated)

Pros (most emphasized)

  • AI-friendly structure (markdown + vault folder + linking)
  • Flexible storage (local-first) with an encrypted sync option
  • Knowledge graph + search + embeddings support rich data/relationships
  • Scales up from personal notes to multi-agent databases to AI-managed wikis

Cons (mostly implied)

  • Power users get the most value—setup and maintenance for AI agents/models/tooling is non-trivial, especially in Level 2–3.
  • Level 1 still requires you to write/populate notes yourself.

Ratings / numerical scores

  • No explicit star ratings or numeric scores for Obsidian.
  • Numbers mentioned appear contextual (e.g., estimated downloads “5 to 10 million”, inbox size 17,000, and “set up in less than 5 minutes” for LLM wiki).
  • No formal product rating is given.

Speakers / perspectives

  • Only one primary speaker is present in the subtitles: Tina Huang (narrator/demonstrator).
  • No distinct secondary speakers are quoted beyond references to external creators/figures (e.g., Tiago Forte, Andrej Karpathy) and AI tools/models.

Overall verdict / recommendation

Recommended: Yes—especially if you want to build AI-powered knowledge systems (AI second brain, AI database, or LLM wiki).

Why: The video argues Obsidian’s local markdown vault structure, linking/search/embeds, and encrypted syncing make it a strong “base layer” for AI agents.

Caveat: If you don’t plan to use AI integrations and only want basic note-taking, the video suggests Obsidian may be overkill compared to simpler tools.

Original video