All video notes
Video notes · Notes and PKM

How To Build The ULTIMATE AI Second Brain (Obsidian + Claude Code)

Video by KJ Rainey on YouTube · 59 min · uploaded April 2026

Notes and analysis by BrainTube, written from the video. Not affiliated with KJ Rainey. Watch on YouTube

In short

The video teaches a two-layer Obsidian vault for Claude Code: a top folder where Claude helps you plan with your goals, notes and journals, and a separate vault per project where Claude only sees that project's files and skills. The scoping idea is sound and the setup steps are concrete. It glosses over permissions, privacy and what happens when the vault grows, and about eight minutes are a demo of the creator's own app.

The argument

Rainey's argument is that an assistant is a poor thinking partner because it knows nothing about you, and a poor worker because it knows too much at once. His fix is to split the work by folder. The top layer holds goals, journals, book and course notes, reviews and every project, and its CLAUDE.md tells Claude to help decide what to do, not to do it. Each project is opened as its own Obsidian vault with its own CLAUDE.md, an inputs, process, outputs and feedback structure, and skills stored as markdown files inside the project. Claude then runs inside Obsidian through a community plugin, so planning and doing happen next to the files. The rest of the video shows this structure running his app, a client email retainer, two YouTube channels and a consulting practice.

Key ideas, with timestamps

  1. 0:41

    A general and its soldiers

    The whole system rests on one metaphor. Layer one is Claude as a strategist that sees your whole life, layer two is Claude as a worker that sees one project. The useful part is the separation of roles, not the military framing.

  2. 2:43

    CLAUDE.md as per-folder instructions

    He explains CLAUDE.md as a plain instructions file that Claude reads in whatever folder you open. In the top layer it holds who you are, your goals, strengths and weaknesses, and the rule that Claude is a thinking partner. Two starter skills ship with it: one that interviews you to create a new project, and a weekly update that refreshes context across the vault.

  3. 4:47

    What layer one actually contains

    His real vault has a goals file, book notes pulled in with Readwise or from Kindle highlights, course and video notes, journals, a long-term strategy file, reviews (he admits he rarely does them) and all project folders. Because Claude can read across every project, he can ask it questions that span his whole workload.

  4. 6:49

    Projects as inputs, process, outputs, feedback

    Each project has the same four parts. For his main YouTube channel, inputs are video ideas, process is research, titles, thumbnails and scripting, output is the published video, and feedback is metrics, comments and his own notes. At 9:34 he opens the project folder as its own vault, so Claude in that window knows nothing else.

  5. 15:00

    Installing Claude Code inside Obsidian

    After installing Claude Code in the terminal at 12:17, he adds a community plugin called Claudian that puts Claude in Obsidian's sidebar. He asks Claude to review the plugin's GitHub repo for anything malicious before installing it, and notes it was clean when he recorded but could change. At 18:26 he picks Opus, sets effort to medium and turns on a mode that skips all permission prompts.

  6. 19:07

    Skills live in the vault, not globally

    He keeps skills as markdown files in a skills folder inside each project instead of installing them into Claude globally. His reasons are practical: his team gets the same skills through sync, skills are easier to edit, and a finance skill never shows up in his YouTube work. The setup skill then interviews you in six rounds, and he adds rules such as tagging every file Claude creates and never editing his own writing without asking.

  7. 23:11

    Write your goals yourself

    The template leaves the goals file empty on purpose. He tells viewers to write their goals outside the AI and hand them over, rather than let the model decide direction. He then runs the new project skill for a finances project that will turn monthly bank statement PDFs into a budget report.

  8. 54:41

    Automate only where quality holds

    His closing tips are the most transferable part. He says "The market only cares about outputs, not processes" and argues you should stop automating any step where quality drops. His example: AI did the research for this video's thumbnail so he could spend about eight hours designing it by hand.

What it gets right

Scoping context by folder is the strongest idea here. A project vault with a short CLAUDE.md and only its own files gives Claude less to misread and less to wander into. Many vault tutorials point the assistant at everything and hope; this one decides what each session should see.

Keeping skills as markdown inside the project is also a good call. They are versioned with the work, shared through the same sync as the notes, and readable by anyone on the team. The habit of asking Claude to review a plugin repo before installing it is worth copying, though a model review is a first pass, not an audit.

The insistence that you write your own goals, and the outputs-over-process rule at 54:41, are sound advice that most setup videos skip.

What it leaves out

  • Permissions. He turns off all permission prompts in the same video where Claude will read bank statements and write across a team-synced vault. A wrong edit then syncs to everyone. Keep prompts on, or at least keep the vault under git or another backup so changes can be reverted.
  • Privacy. Financial PDFs, client call transcripts and journals are sent to the model provider when Claude reads them. That may be fine for you, but it is a choice the video never mentions.
  • Scale. Layer one "has access to all of it", but Claude Code does not load a whole vault at once. It searches and opens files as it needs them, mostly by filename and text match. As the vault grows, recall depends on how well notes are named and linked, and on the weekly update skill actually being run.
  • Overstated autonomy. The intro promises work done while you sleep. Nothing in the video runs on a schedule. Dispatch from the phone needs the computer left on, and he calls it glitchy at 32:04.
  • The app segment. From 34:07 to about 39:36 he demos Signal, his own paid YouTube learning app. It is a product tour, not part of the method.

How this applies to memory across AI tools

The design is tied to one assistant on one machine. The vault works because Claude Code is running inside that folder. Open ChatGPT or Cursor and none of these notes, goals or project files are there. The structure is portable because it is plain markdown, but the memory is not, until something exposes it to other tools. See persistent AI memory across tools for the general problem.

The usual fix is an MCP server over the knowledge you want every assistant to share. BrainTube takes that route for what you watch and read: 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. A sensible split is to keep project files and skills in the vault, as Rainey does, and put reference material in a layer every tool can query. The trade-offs are covered in this comparison with Obsidian.

Who should watch it

Watch it if you already use Obsidian or plain markdown folders and want a concrete pattern for running Claude Code against them. The method is explained from 0:41 to 10:15, and the step-by-step setup runs from 10:55 to 29:19; the project-template trick at 26:36 and the skills discussion at 19:07 are the most reusable parts. The closing tips from 54:41 are worth five minutes on their own. Skip 34:07 to 39:36 (his app demo) and skim the use cases from 39:36 onward unless you run a content business. Developers who already use CLAUDE.md files will find little new beyond the per-project vault idea.

Frequently asked

Can I use Claude Code with an Obsidian vault?
Yes. An Obsidian vault is a normal folder of markdown files, so you can run Claude Code in that folder from the terminal and it can read and edit the notes. In this video a community plugin called Claudian puts Claude Code in Obsidian's sidebar, but he also notes you can just open the folder as a project in the Claude desktop app.
What should go in a CLAUDE.md file for a notes vault?
Keep it short: who you are, what the folder is for, what Claude's role is, and rules it must follow. Rainey's top-level file says Claude is a thinking partner, while each project file states the project's goal and Claude's job in it. Good rules include tagging files the AI creates and never editing your own writing without asking.
Is it safe to let Claude edit my Obsidian notes?
It is reasonably safe if you keep permission prompts on and have a way to undo changes, such as git or regular backups. Turning off permissions, as done in this video, means mistakes are applied without review and, in a synced vault, spread to everyone. Also remember that any file Claude reads is sent to the model provider.

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.