All video notes
Video notes · Notes and PKM

Obsidian + AI: How to Do It The Right Way (Claude Code + Obsidian)

Video by Linking Your Thinking with Nick Milo on YouTube · 14 min · uploaded September 2025

Notes and analysis by BrainTube, written from the video. Not affiliated with Linking Your Thinking with Nick Milo. Watch on YouTube

In short

Nick Milo argues that AI should be kept out of your own notes but used in narrow ways around them, and shows Claude Code reading his Obsidian vault and writing its output into a separate AI-only vault. The separate vault is a smart, cheap rule worth copying. His privacy framing is too generous: Claude Code sends your note contents to a cloud model, so the setup is only as private as your account's data settings.

The argument

Milo starts from the view that an Obsidian vault is a private thinking space, and that letting AI write into it dilutes your own voice. Still, AI is good at things people are bad at, like scanning years of notes for every mention of a theme. His answer is a "barbell": on defence, avoid over-generating and pick a privacy level on purpose; on offence, use AI for reflection and research, but keep everything it produces in a dedicated AI zone. You only move something into your real notes by hand, which creates what he calls good friction. Claude Code fits because Obsidian is just local markdown files, so an agent can read, edit and restructure them directly. He closes with Obsidian CEO Steph Ango's position that AI is not on the roadmap and would have to be private if it ever arrived.

Key ideas, with timestamps

  1. 0:31

    Should AI touch your notes at all?

    He opens with the case against: Obsidian is loved for being offline and private, and he sees his vault, which he calls his Ideaverse, as a space for his own thinking. AI brings other people's words into it. He then concedes AI is better at searching every past mention of a topic or finding a passage in a long podcast.

  2. 1:32

    Imagine, discern, integrate

    His IDI framework: let AI widen what you think is possible, judge what it gives you (it may be wrong but still useful), then connect the useful parts to your work and life. Discernment is the step he leans on later when the AI's analysis gets details wrong.

  3. 2:35

    Defence: over-generation and a privacy policy

    Do not let generated text crowd out your own writing. Then pick a spot on a privacy spectrum: fully local at one end, a chat app that trains on your data at the other, and cloud tools that do not train on your data in the middle. He places Claude Code in the middle.

  4. 3:36

    Offence: reflection, research and an AI zone

    AI is good at reflecting the last 30 days of your thinking back to you and at starting deep research. His key rule is a dedicated place for AI output, so nothing it writes reaches your own notes unless you choose to copy it in.

  5. 4:39

    Why Claude Code fits Obsidian

    Claude Code works on files and folders, and an Obsidian vault is files and folders. That means it can analyse notes, edit across hundreds of them, or launch sub-agents to research the web. He warns to back up the vault first, and points out the notes stay in plain markdown you own.

  6. 5:40

    Demo one: a 45-day reflection

    He runs Claude in the terminal, points it at the vault and asks for an analysis of the past 45 days of notes. It writes a report of almost 1,000 words into his separate AI vault, covering book writing, business, routines and his producer versus creative modes. He notes one theme about Montana and LA is not quite accurate.

  7. 7:14

    Demos two and three: word patterns and metadata

    He asks it to find every use of "Ideaverse" and analyse the pattern, and it reports a count in the thousands peaking in the last month, which he says he still needs to verify. Then, by dictation, he has it fetch a photo from the web for every person note and fill the image field.

  8. 10:22

    Obsidian's own position on AI

    As of the September 2025 upload, there is no built-in AI and it is not on the roadmap; AI comes through community plugins or external tools. He plays clips of Steph Ango saying Obsidian will not join the rush to add AI buttons everywhere, and that any future AI would have to be private.

What it gets right

The separate AI vault is the best idea here. Writing every AI output into its own vault (or at least its own folder) costs nothing and solves a real problem: after a few months of agent use, many vaults become a mix of your notes and generated text that you cannot tell apart. Making yourself copy things across by hand forces a judgement on each one.

Discernment in practice. He shows the AI getting something wrong in the 45-day reflection and says so on camera. Treating agent analysis as a draft to check, not a finding, is the right habit, and few demo videos model it.

Files you own are the right base. Because the vault is plain markdown, any agent or tool can read it, and you can switch tools without exporting. That portability matters more than any single AI feature.

What it gets wrong or skips

Claude Code is not private in the way Obsidian is. The notes stay on your disk, but every file Claude Code reads is sent to Anthropic's servers for the model to process. Whether that data can be used for training depends on your account type and settings, and Anthropic changed its consumer terms in 2025 to ask users to choose. Placing it in the middle of his spectrum is fair only if you have checked that setting. A truly local setup needs a local model, which he does not show.

Unverified numbers go straight into the AI vault. The Ideaverse count is reported and saved before anyone checks it. Asking the agent to show how it counted (for example, the search it ran) is a quick way to make discernment concrete.

Bulk edits deserve more care. The image task writes into every person note in his real vault, which breaks his own AI-zone rule. It is also easy for a web search to return the wrong person with the same name. Running it on a copy, or under version control, and spot-checking results would be safer.

Several plugs. The video includes a workshop pitch (9:49) and an archetype quiz (13:00). The method itself takes about eight minutes.

How this applies to memory across AI tools

Milo's setup is one agent reading one folder. It works well in the terminal, but your vault is invisible to the assistant you use in a browser or in Cursor. The same split he recommends, your own words in one place and machine-made material in another, also applies to what you collect from outside: transcripts, articles and papers are useful to search but are not your thinking.

One option is to keep that outside material in a store any assistant can reach. BrainTube does this: it saves YouTube videos with their transcripts, articles, PDFs and podcasts into one compiled knowledge base and exposes it as an MCP server, so Claude, ChatGPT, Cursor and other MCP clients can search it, while your Obsidian vault stays your own writing.

For a direct comparison, see a compiled knowledge base vs Obsidian, and for the privacy side, AI memory portability.

Who should watch it

Watch it if you use Obsidian and are wary of AI taking over your notes, or if you want to see what Claude Code does with a real vault. The barbell and AI-zone rule from 2:35 to 4:08 is the part to take away, and the demos from 5:40 to 9:19 show what a first session looks like. Skip it if you want a technical setup guide (there is no configuration, no CLAUDE.md and no plugin walkthrough), and skip 9:49 to 10:22 and 13:00 onwards, which are promotions.

Frequently asked

Can Claude Code work with an Obsidian vault?
Yes. An Obsidian vault is a folder of markdown files, so Claude Code can be started in that folder and asked to read, search, edit or reorganise notes. Back up the vault or put it under version control first, because the agent can change many files at once.
Does Obsidian have built-in AI?
As of September 2025, when this video was published, Obsidian had no built-in AI and its CEO said it was not on the roadmap. AI features come from community plugins or external tools such as Claude Code. Check Obsidian's current roadmap, since this may change.
Is using Claude Code with Obsidian private?
Only partly. Your files stay on your computer, but the contents Claude Code reads are sent to Anthropic's servers to be processed. Whether they can be used for training depends on your plan and privacy settings, so check those before pointing an agent at personal notes.

Keep videos like this one searchable

BrainTube saves YouTube videos with their transcripts into one compiled knowledge base that Claude, ChatGPT, Cursor and other MCP clients can search.

Start free

Go deeper

More video notes

Are you the creator and want this page changed or removed? Contact us and we will act within 48 hours.