How I Use Obsidian + Claude Cowork to Run My Life
Video by Linking Your Thinking with Nick Milo on YouTube · 19 min · uploaded June 2026
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 shows how he points Claude Cowork at his Obsidian vault through a small translation layer: a me.md identity file, a vault map, a skill map and a one-line CLAUDE.md that redirects to me.md. The core idea, keep your identity, maps and skills as plain files you own and rent the AI on top, is right and easy to copy. The portability is less complete than he suggests, the privacy claim needs checking against current settings, and a fair part of the video promotes his paid product.
The argument
Milo's argument is that your notes and AI workflows should be owned and the AI tool should be rented. He builds a three-layer setup. The inner layer is his Obsidian vault, plain Markdown files organized into Atlas, Calendar and Efforts. The middle layer is an AIOS folder holding the files that tell any assistant who he is, how the vault is laid out and which skills exist. The outer layer is whichever AI tool he plugs in, currently Claude Cowork. Because a large vault is impossible for an assistant to scan, the maps do the navigation, and because the files are tool-neutral, he says he could switch to Codex or a local model without rebuilding anything.
Key ideas, with timestamps
- 1:00
The vault is just a folder
He stresses that an Obsidian vault is a folder of Markdown text files that any editor, from TextEdit to VS Code, can open. His top-level folders follow his ACE scheme: Atlas for ideas and reference, Calendar for time-based notes like journals and meetings, and Efforts for projects and tasks. Plain files are what make the rest of the setup portable.
- 3:15
Pointing Cowork at the vault
In the Claude desktop app, Cowork mode is granted access to one local folder. Claude can then read, edit, move, rename and create files there without uploads. He points it at a parent knowledge folder that contains his main vault, called Ideaverse, plus other vaults.
- 4:30
Why Claude, for now
He picks Claude because, in his words, it does not train on your data, keeps data on its servers for a rolling 30 days and is less restrictive than other frontier apps. He uses Sonnet by default, Opus for heavy synthesis and Haiku only for simple web-clipper summaries. He adds that this may change, and that he expects a local model to replace it eventually.
- 6:30
17,000 notes need a translation layer
Pointing an assistant at about 17,000 notes without guidance is, he says, a bad idea: it will sample a few files and act as if it read them all. His fix is a separate AIOS folder for maps, manuals and anything AI generates, so his own notes stay clean and AI output can be cleared out quickly.
- 7:30
me.md and a one-line CLAUDE.md
me.md describes who he is, how he thinks and how he wants an assistant to work with him. The CLAUDE.md that Claude reads at session start contains only one instruction: go read me.md. Keeping CLAUDE.md nearly empty is his way of avoiding lock-in to one vendor's file.
- 9:30
Vault map and skill map
The vault map is a table of contents and manual that tells the assistant where things live, so it can load only relevant notes. The skill map lists each skill, what it does and when to use it. He insists skills are stored in his vault, not inside Claude, even when Claude helps write them.
- 11:45
The session-start prompt
He does not trust automatic loading, so every session begins with a short prompt: read me.md, review the vault map and skill map, confirm, then wait. He fires it with a TextExpander shortcut. It is a practical admission that models often start without the context you expect.
- 13:00
The skills in use
A scheduled daily brief runs at 6:00 a.m., pulling from Gmail, his calendar, ClickUp and the vault, and he leaves replies for the assistant in curly brackets. Other systems include a daily log, a weekly review he reads only after doing his own, a rock tumbler that gives feedback as open questions, a chronicler that saves a chat verbatim on request, a janitor for maintenance and a courier that strips personal details before sharing a note with his team.
What it gets right
Own the files, rent the model. This is the most important idea in the video, and it applies well beyond Obsidian. Identity, instructions and skills written as plain Markdown in a folder you control survive a change of tool. The one-line CLAUDE.md pointer is a neat trick. A similar effect comes from AGENTS.md, a plain instruction file that several coding agents read.
Maps beat brute force. His point at 9:45 is correct and often ignored. An agent with file access does not read 17,000 notes; it greps a few and summarizes. A maintained index tells it where to look and makes its answers checkable.
Keep AI output separate. A dedicated AIOS folder for generated content means you can always tell your writing from the assistant's, and delete the latter without risk. Few people building these setups think about this.
He keeps his hand on the wheel. Doing his own weekly review before reading the AI version, and using the rock tumbler for questions rather than rewrites, is a sensible line between help and outsourcing your thinking.
What it skips or overstates
Portability is partial. The files travel; the behavior does not. The 6:00 a.m. brief runs as a scheduled task inside Claude, the connectors to Gmail and ClickUp are set up in Claude, and the skill format follows Anthropic's conventions. Moving to another tool means rebuilding those parts.
Check the privacy claim yourself. Anthropic's consumer data terms have changed over time, and whether chats can be used for training depends on your account settings. Before pointing any assistant at years of journals, check the current policy and your own setting rather than relying on a video.
Maps go stale. A hand-written vault map is only useful while it matches the vault. He mentions a janitor system but does not show how it keeps the maps accurate. Search over meaning, not just file names, would cover what a map misses.
Write access plus email is a risk. An assistant that reads Gmail and clipped web pages and can also edit and move files in your vault can be steered by text inside those sources. Keep the assistant's write access narrow, and keep backups or version control on the vault.
Promotion. From about 15:00, and again at 18:30, much of the detail is held back for his paid product. The free part is still enough to copy the three core files.
How this applies to memory across AI tools
Milo's layer covers what you write. It does not cover what you watched, heard and read but never typed up, which for many people is most of what they learn. That material needs the same rule he applies to notes: keep it in one place you control and let any assistant reach it.
BrainTube handles that side. 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 next to a notes vault like his.
For how this compares with an Obsidian-only setup, see the comparison with Obsidian. For keeping one memory across assistants, see one memory across ChatGPT, Claude and Cursor.
Who should watch it
Watch it if you already keep notes in Obsidian and want to connect an assistant without dumping your whole vault into it. The essentials are 6:30 to 12:45, where he builds the AIOS folder, me.md, the vault map and skill map and the session-start prompt. Watch 13:00 to 18:00 for ideas on which skills are worth building. Skip it if you are new to Obsidian; he sends basics to another video. Skip from 18:15 if you do not want the product pitch.
Frequently asked
- What is a me.md file?
- In Nick Milo's setup, me.md is a plain Markdown file describing who you are, how you think and how you want an AI assistant to work with you. It replaces putting that information in a vendor-specific file. His CLAUDE.md contains only one line that tells Claude to read me.md.
- How do you connect Claude Cowork to an Obsidian vault?
- Open the Claude desktop app, switch to Cowork mode and grant it access to the folder that holds your vault. Claude can then read and edit the Markdown files directly. Milo adds a vault map and a session-start prompt so it reads the right notes instead of guessing.
- Can AI read a large Obsidian vault?
- An assistant with file access can open any note, but it will not read thousands of them in one session. It samples a few and may act as if it covered more. A maintained index or vault map, or search over the content, tells it which notes are relevant.
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.
Go deeper
More video notes
- Build an Obsidian BACKEND Not a Second Brain! · Eric Michaud
- Complete Guide to building an AI Second Brain with Claude · Ray Fu
- The NEW Way to Build a Second Brain w/ Claude (2027) · Nick Saraev
- Build a Self Improving Claude Knowledge Base · Paul J Lipsky
Are you the creator and want this page changed or removed? Contact us and we will act within 48 hours.