LLM Wiki in Obsidian: A Step-by-Step Setup Guide (No Code)

This guide walks you through building an LLM Wiki inside Obsidian from zero — no terminal, no Python scripts, no developer background. You will set up a folder structure, install three plugins, write a one-page schema, and run your first compile through Claude's web UI. By the end you will have a working second brain that grows every week instead of rotting in a Notes folder. Total time: about 45 minutes. If the LLM Wiki concept still feels abstract, start with our plain-English primer — then come back here for the hands-on version.

Obsidian is the best home for an LLM Wiki because it speaks markdown natively, it has wikilinks and backlinks built in, and you can see what the LLM wrote as a living document rather than a query log. Karpathy himself built his LLM Wiki in Obsidian. Let's build yours.

What you will need

• Obsidian (free) — download here
• Claude Pro or Claude API access — free tier works for tiny wikis; Pro is better for real use
• 15 MB of disk space — yes, really, that is the whole infrastructure budget
• Zero terminal commands — everything happens in Obsidian's UI or the Claude web app

That is the entire stack. No vector database, no embedding pipeline, no Docker. If you have ever been intimidated by an LLM Wiki "quick start" that opens with git clone, you will like this one.

Step 1 — Create the Obsidian vault

Open Obsidian and click Create new vault. Name it whatever you want — I will use llm-wiki for this walkthrough. Pick a location you can find later, like ~/Documents/llm-wiki on macOS or Documents\llm-wiki on Windows.

Inside the vault, create three folders using Obsidian's right-click menu on the sidebar:

• raw/ — this is where you will drop raw sources (PDFs, articles, transcripts, screenshots)
• wiki/ — this is where the LLM will write the compiled output
• meta/ — this is where your schema and compile log will live

If you have used Obsidian before, this structure will feel familiar. The only trick is that raw/ holds the input and wiki/ holds the output, and you never manually edit wiki/ files.

Step 2 — Install the three plugins

Open Settings → Community plugins → Browse. Install these three, in order:

1. Dataview — lets you query your vault like a database. You will use it to build an index page that auto-updates as the wiki grows.
2. Templater — lets you save schema templates and reapply them. Handy for iterating on your schema without copy-paste.
3. Obsidian Git — automatically commits every compile. This is your safety net; when an LLM compile goes wrong, git checkout gets you back to yesterday.

After installing each one, go to Settings → Community plugins → [plugin name] and turn it on. Obsidian Git needs one extra step: set the auto-commit interval to 30 minutes so every session is automatically saved.

Step 3 — Write your schema

This is the single most important step. Your schema tells the LLM how to write the wiki. A good LLM Wiki schema is one page long, a bad schema is five pages long, and the wrong schema is no schema at all.

Create meta/schema.md with something like:

# Schema: Personal LLM Wiki

## Entity types
- **concept**: an idea, technique, or abstraction
- **person**: someone whose work I reference
- **tool**: software, library, or service I use
- **source**: a paper, article, video, or book I read

## Page format
Every wiki page starts with frontmatter:

tags: [entity-type, topic]
created: YYYY-MM-DD
updated: YYYY-MM-DD

Then the body in this order:
1. Summary (1-2 sentences, what this is)
2. Key points (3-5 bullets)
3. Related (wikilinks to other pages)
4. Open questions (things I have not yet resolved)
5. Source refs (links to raw/ files)

## Rules
- Use [[wikilinks]] for anything that is itself a page
- Never overwrite Open Questions without flagging
- When two sources contradict, note it in Open Questions
- Keep each page under 500 words

Your LLM Wiki schema will evolve. Start simple, compile a few sources, and then adjust the rules when the output disappoints you. If you want a shortcut, our LLM Wiki templates page has five pre-written schemas for general, research, engineering, product, and SEO use cases — each one battle-tested on real projects.

Step 4 — Your first compile

Find one source to start your LLM Wiki with. A PDF you recently read, a YouTube transcript, a substack article — anything you actually want to remember. Drop it into raw/ (Obsidian will copy the file into the vault automatically if you drag it in).

Open the Claude web app to run your first LLM Wiki compile. Start a new chat and paste the contents of meta/schema.md at the top. Then paste the contents of your raw source below it. Finally, add this prompt at the end:

Following the schema above, draft the wiki pages that should be created or updated from this source. Output each page as a separate markdown block with its full filename. Do not overwrite Open Questions. Keep each page under 500 words. Flag any contradictions with existing knowledge.

Claude will return a set of markdown blocks. Each block is one wiki page. Copy each one into a new file in your wiki/ folder. That is your first LLM Wiki compile.

If the LLM Wiki output disappoints you, do not edit the wiki files — edit the schema. The first ten compiles are really about tuning your schema until the LLM writes the way you want it to write.

Step 5 — Build the index page

Create your LLM Wiki's index.md at the root of the vault. Drop in this Dataview block:

\`\`\`dataview
TABLE file.mtime as "Updated", tags as "Type"
FROM "wiki"
SORT file.mtime DESC
\`\`\`

Dataview will now auto-render a live table of every wiki page, sorted by most recent. This is your front door. Whenever you open the LLM Wiki vault, index.md shows you what the wiki knows.

Add a ## Recent open questions section below the table using another Dataview query that pulls any page with an Open Questions heading. This is how you catch contradictions early — anything flagged during compile shows up here.

Step 6 — Build the compile loop into your routine

The magic of an LLM Wiki happens when you run the compile weekly, not once. Here is the rhythm that works for most people:

1. Drop in new sources as you find them. Articles, papers, notes from meetings — just drag them into raw/. Do not try to process them yet.
2. Once a week (Sunday mornings work well), run the compile. Open Claude, paste the schema, list the new files in raw/, ask for the wiki updates. 10-20 minutes.
3. Read the new wiki pages. If anything feels wrong, fix the schema, not the page.
4. Commit. Obsidian Git will auto-commit if you set it up in step 2, but a manual commit after the compile is a nice checkpoint.

After a month, your wiki will have 30-50 pages and will start to feel like an actual second brain. After three months, your LLM Wiki will cover most of what you have been thinking about. That is when the compounding kicks in — every new source updates five existing pages instead of creating orphans, and suddenly you have a real reference work.

Common pitfalls

• Overwriting the LLM's output manually. Stop. Edit the schema instead.
• Schema bloat. If your schema is over two pages, throw half of it out. Simpler schemas produce cleaner wikis.
• Not reading the LLM Wiki output. If you never re-read what the LLM wrote, you will not notice when the schema is failing you.
• Forgetting git. One bad compile can overwrite good pages. Git is your safety net.
• Trying to compile everything at once. Do one source at a time in the first month of running your LLM Wiki. You are tuning the schema, not processing a backlog.

Where to go next

• New to the pattern? Read the LLM Wiki primer for the conceptual background.
• The full writer track: LLM Wiki for Writers — every aillm.wiki resource written for non-developer knowledge workers running this pattern through Obsidian, Notion, or a browser tab.
• Want to understand the RAG debate? See LLM Wiki vs RAG for the decision framework.
• No Obsidian? The no-code LLM Wiki tutorial runs the same pattern in ChatGPT or Notion.
• Skip the schema-writing weekend. The LLM Wiki Starter Kit ships 5 tuned CLAUDE.md and schema.md files plus an ingest pipeline. A drop-in Obsidian Vault is on the roadmap — Starter Kit subscribers hear first when it opens.

If you prefer to build it yourself, join our occasional newsletter — we email when we have new LLM Wiki resources, schema patterns, or community discussions worth your time. No fixed schedule, unsubscribe in one click.