Speaker A
This is my AI OS, Obsidian, Claude, and all the pieces that you need to bring them together. By the end of this video, you'll know how I use this system, the actual AI-enhanced skills I use, and how
Nick Milo explains how he uses Obsidian and Claude Co-work to create a flexible AI-enhanced system for managing ideas and productivity.
Key Takeaways
- Using Obsidian as a markdown-based note system ensures data portability and longevity.
- Linking notes in Obsidian builds a dynamic network of ideas that grows stronger with use.
- Claude Co-work enables seamless AI interaction with local files while respecting user privacy.
- A layered AI OS approach separates personal knowledge, AI interaction protocols, and AI tools for flexibility.
- Futureproofing your AI workflows means owning your data and being able to switch AI providers easily.
Summary
- Nick Milo introduces his AI operating system combining Obsidian, Claude Co-work, and other AI tools.
- The system has three layers: ideaverse (personal thoughts), core maps/manuals for AI interaction, and external AI tools.
- Obsidian is used as the foundation, storing notes in markdown files within folders called vaults.
- Primary Obsidian folders include Atlas (knowledge), Calendar (time-based notes), and Efforts (projects and tasks).
- Obsidian’s unique linking feature helps visualize and strengthen connections between ideas.
- Claude Co-work is used as the AI tool to interact with Obsidian files without needing repeated uploads or context explanations.
- Claude is preferred for privacy (no training on user data) and flexibility compared to other AI ecosystems.
- Different Claude models (Sonnet, Opus, Haiku) are used depending on task complexity and resource needs.
- The system is designed to be futureproof, allowing easy switching between AI tools without losing data or workflows.
- A translation layer with maps and manuals is created in Obsidian to help AI navigate large vaults effectively.
Full Transcript — Download SRT & Markdown
Speaker A
you can set up your own system that's both futureproof and works with any AI tool instead of locking you into Claude, Codex, Gemini, or Frank. Hi, I'm Nick Milo, and yes, I'm a real person, and I help you create an idea, a home
Speaker A
for capturing and using your best thinking. My AI operating system has three layers. The first one is my ideaverse, which is my own thoughts.
Speaker A
Second, the core maps and manuals that allow my AI or any external AI to consistently and safely interact with my ideas so that we can communicate back and forth. And the third layer is whatever external AI tools or
Speaker A
applications I'm plugging into. And in this case, it'll be Claude Co-work. The best part is if Claude goes away tomorrow, I could instantly swap it out with Codex, with an open source model, with whatever I need. I have all my
Speaker A
files, skills, and AI core documents built to go with me, ready to use in whatever AI ecosystem or environment I want to connect it to in the future. And that means we've got to start with your ideaverse in Obsidian.
Speaker A
Obsidian is at its core just a folder of notes sitting on your computer. The main folder is called your vault, and that's basically it. Inside that folder, your notes are individual text files written in a format called markdown. You can
Speaker A
open those files in Obsidian and they look beautiful, but you can also open them up in any other app that reads markdown. So that could be TextEdit, it could be VS Code, IDEA Rider, and even AI tools like Claude Co-work. So even if
Speaker A
Obsidian disappears tomorrow, as long as there are computers, you'll have a way to open your notes. And this is essential within Obsidian. My primary folders are Atlas, Calendar, and Efforts, or ACE for short. Atlas is for your knowledge, your ideas, your
Speaker A
reference material, the things that you want to think about and be able to build and develop over time. Calendar is for your time-based notes, daily notes, journals, meetings, anything anchored to specific moments. And Efforts are for the things that you're working on, your
Speaker A
projects, your big, big efforts, your tasks, works, anything productivity-oriented. One of the things that made Obsidian so unique in the early days was the ability to link notes to each other.
Speaker A
All you have to do here is type double bracket twice and then start typing the note title or a keyword, and you'll see that there are all these possible connections you can create. And as you make those connections, you can watch
Speaker A
your graph view grow. Individual notes become these dots, and then links become lines, and suddenly you can see the surprising amount of relationships between these links and dots, between our ideas that you never would have noticed otherwise. Each time you revisit a
Speaker A
linked note connecting to another note, you're strengthening the connection between them, both in the app, but also in your own mind. And in this way, this system that we're creating gets better the more that you use it. If you
Speaker A
want the full details on how to set up the basics or what to do if you want to go more advanced, check out the description link below. It covers all the Obsidian basics so you're not lost as we go deeper. In this way, Obsidian
Speaker A
organizes the heart of my AI operating system, which has to be my own thoughts.
Speaker A
Now, we're going to jump from the innermost layer to the outermost, the AI toolkit that will connect to your ideaverse. So, for this, we'll be using Claude Co-work.
Speaker A
To use Co-work, download the Claude desktop app and then sign in with your Claude account. Inside the desktop app, you'll see a toggle bar that lets you switch between three modes. Co-work is the one in the middle. It works when you
Speaker A
select a folder on your computer, you click allow, and then just like that, Co-work can see everything in the folder that you've linked. You don't have to keep uploading documents or reexplaining context. Claude can now read files, make
Speaker A
modifications, move things around, rename them, and even create new files on your behalf. Now, let's talk about what we're actually going to give Claude access to. Here's how my Obsidian is currently set up. And again, just files and folders on my computer. You can see
Speaker A
on a Mac I have my Macintosh hard drive, users, and then the Nick folder. I just simply then go to knowledge and then I open that up, and these are the different Obsidian folders that are vaults that I
Speaker A
can open up independently. Where I spend 99% of my time is in this idea folder.
Speaker A
Also, I spend time in some other vaults as well. So what I do is I go into Claude, click on Co-work if I'm not there already, point Claude to the right folder. For me, I'm just going to simply point it at this knowledge folder. You
Speaker A
can see I've pointed it at other things in the past, but my knowledge folder contains my main vault, which for me is simply called Ideaverse. Now, why am I using Claude specifically? There are a few reasons for now. The first is that
Speaker A
Claude says that they don't train on your data. That is a non-negotiable for me, especially when I'm pointing it at my personal notes, my journals, and years of private thinking. They've also been transparent that data lives on their
Speaker A
servers for a rolling 30-day window, which feels like a reasonable trade-off for having access to frontier AI models.
Speaker A
The other reason I'm using Claude is that it's one of the least restrictive frontier apps out there. Other tools are actively trying to entrench you. They want you to build your habits, your workflows, your dependencies inside their ecosystem so that you can't leave.
Speaker A
Now, Claude is better about this than most, and this might always change in the future. So, let's just have that little caveat. You might be wondering which models do I use in Co-work. Well, the best general approach and to allow
Speaker A
you to use Pro more often, try to use the Sonnet model first. That's the middle model. It's less intensive, but it's good for most tasks. When something is meteor, a complex synthesis, a deep research task, something where the
Speaker A
quality really matters, that's when I'll switch to Opus. It uses tokens faster, but it does the best job possible right now. The lowest model is called Haiku. I rarely use that one unless I'm using the Obsidian Web Clipper, and then I'll just
Speaker A
have it download and summarize articles on my behalf. But it's really only for those types of very simple, lightweight tasks. Otherwise, I end up double-checking its work, and then what's the purpose? The real goal is that you want to own your ideas and thinking and
Speaker A
your AI workflows, and then you just want to rent the actual AI tool itself. ChatGPT, that was the default not long ago, and today it's Claude. Okay, great. Do we really think it's going to be one of
Speaker A
those two in two years? It might be, but it might just as well be something else.
Speaker A
And as long as your files can travel with you, your AI can just be a layer that you swap out eventually to a local model, I hope, which is where I think all this is going eventually, probably powered by Apple's silicon on device.
Speaker A
But what do I know? Now, when you're ready to go extra deep to set up Cloud Co, well, for a system like this one, check out the details in the description below. So, now here's the thing. If you
Speaker A
point Co-work at Obsidian and let it do its thing, that might be a horrible idea. My Obsidian vault is around 17,000 notes. Doesn't matter how many notes you have, but when you get to that amount, you point AI at something like that, and
Speaker A
it has no feasible way to navigate. Which is exactly why we need to set up a translation layer, the middle layer of the AI operating system. And this is where our maps and manuals live.
Speaker A
All right. So, what we're going to do is we're going to go into Obsidian and create a folder called AIOS. This is a separate area from your ideaverse that lives inside your mai
Speaker A
important. And here's why. If I'm generating AI content, I want to always be able to quickly isolate it, clear it out when necessary, and that way always keep my thoughts and ideas in my idea verse very clean and clear. Inside this
Speaker A
folder, we're going to create a few of our main maps. But the first and most important is the MIMD file. Think of MIMD as your portable identity for AI.
Speaker A
If you used Claude Co-work before, you might know about Claude MD. It's essentially a master file that Claude reads every time you start a new session. But that's Claude specific. So if you switch to a different AI tool
Speaker A
next month, that file doesn't automatically go with you, and it's definitely not tailored to an agnostic AI world. Instead, you create your mi.mmd file, which is here, a plain markdown file, and it tells any AI, in this case, here's who I am. Here's how I
Speaker A
think. Here's how I want you to work with me. And here's how it works in practice. At the root level of the folder that I point Claude to, so for me, it's my knowledge folder. That's where I put my Claude MD file. All it
Speaker A
says is go immediately to me. MD. That's it. When Claude starts a session and it reads that oneline instruction, it goes to me. MD. And now it knows me. I don't want anything else in that claude file because that locks me into that claude
Speaker A
ecosystem. There's a lot more that we're covering about this in the description if you'd like a full guide on how to set up your own Mi.mmd file. You may have noticed that I have a claude.md file in my Ideavor itself. So if we go back to
Speaker A
the finder, if I twirl down Ideavverse here, we can see claude exists here as well. And that's mainly for you just to reduce confusion. If you only have one vault that you're pointing AI towards, you'll want your Claude file right here.
Speaker A
It just so happens that I mine's a little bit more complex. I'm doing a few more things than the average person when it comes to all this sort of knowledge management and AI permissioning. So that's why I have this new knowledge
Speaker A
folder and then my claw.md file is there. So this is the one that my claude is reading. But if you only have a single Obsidian vault that you're using, you'll want to make sure that your Claude file exists right here, just so
Speaker A
Claude knows to immediately go to mi.md. Next up, the vault map. The vault map answers a specific question. How should AI move through your notes? You can think of it as your vault's master table of contents and manual. It lets AI
Speaker A
navigate to the right note or collection of notes and create notes on your behalf without having to scan your entire knowledge base. Because if you point an AI at 17,000 interconnected notes with no map, it's going to pretend that it's
Speaker A
reviewed all of your notes when it's actually just kind of like cherrypicked and sampled and barely touched any of them. And the results that you get won't be accurate. They won't be precise because it doesn't know where to go and
Speaker A
it doesn't know how to get relevancy fast. But give it a map and it can figure out which files are actually relevant. It can isolate those, load them into its context, skip the rest.
Speaker A
The third key map is the skill map. The skill map gives very clear instructions about what skills exist for your AI, what they do, and exactly when AI should use them. And this is where I want to be
Speaker A
really direct about something. Keep your skills here in your own system, in your own notes, and not in claws or any other AI tools. It might be a little bit more convenient when you initially create the skill, but then you're trapped every
Speaker A
day, week, month, and year afterwards. Whenever you create a skill, just have your AI tool of choice, put it right here in your skills subfolder. To be clear, I do use Claude and AI tools to actually create the skills and review
Speaker A
the skills after they are made. That's what AI should be doing on your behalf.
Speaker A
But the skills are stored separately from Claude and documented inside this skill map here. These maps are the translation layer between Claude on the outside or whatever that AI tool is and Obsidian and our ideaverse of notes in
Speaker A
the middle. Without you having this translation layer in the middle, I cannot guarantee that you won't find yourself trapped by the tech lords.
Speaker A
Let's avoid that. Now, let's take this translation layer even further. Once you have your translation layer set up, it's time to point Claude Co-work at your idea verse. Let me do that now by pointing it to my knowledge folder in
Speaker A
which the idea verse is underneath. And now it's going to have access to my MI.MD file, my vault map, and my skill map. The big three files that make this whole thing work. Now, I found that the most reliable way to make sure that
Speaker A
everything is loading correctly is to also fire off a prompt at the start of every new session. Here it is on screen now and you can see first please read the MI.MD file. I'm letting it know exactly where it is by saying idea
Speaker A
verse. It's in the ideiverse folder. Then review vault map and skill map for relevant context. Confirm you've read then await instruction. That's it. Now this isn't 100% necessary, but these Frontier AI models right now, the best models out there, they are brilliant.
Speaker A
But all models suffer from amnesia. And even if you think it knows what you know and has the context that you want it to have, it might not. Because if I start a new task, it's not a guarantee that all
Speaker A
that context will be loaded up. If you can program a hotkey, you can save yourself a lot of time. Set up a quick shortcut in text replacements on Mac or with an app like Typonator, which I'm currently using, or whatever works best
Speaker A
on Windows and Linux, so that you can just with a couple keystrokes paste this whenever you need it. So for me, I use SSK and boom, new chat ready to go. It's going to load that up and then whatever
Speaker A
I say next, I know with confidence I'm going to get accurate results back. So here are a few things that I can do with my AI assistant. Every morning, a daily brief is already waiting for me in Obsidian. Inside this brief, there are a
Speaker A
few things that are incredibly helpful for starting the morning. It gives me a weather report. And this is especially useful right now because I'm in New York City and I know, oh, I need to take an umbrella with me before I head outside.
Speaker A
It's raining all day today. That was really useful. It kept me dry. Big win for the daily brief. It also tells me what happened yesterday, what I worked on, what I moved forward, what's still open. It surfaces any important emails
Speaker A
or logistics that I need to be aware of. It reminds me of upcoming travel and deadlines. It reminds me of the momentum on my big project, the book. What was I doing? What do I need to do? What's
Speaker A
happening in my intellectual life right now? What might I be forgetting? And what are the two or three things that I actually need to do today? Throughout the daily brief, I have open note sections. These are places where I can
Speaker A
leave notes throughout the day to my AI. That way, I can be on the brief when I'm on my phone and then when I come back to the computer here, all my notes are there. I can put them in curly brackets
Speaker A
and AI knows to look at the curly brackets. Other times I just indent in the bullets and I make my notes there.
Speaker A
AI knows to respond to those as well. All from the daily brief. Now this brief, it's generated by one of the most important skills in my AI operating system, the daily brief skill. And it's part of a broader system called the
Speaker A
daily trident system. This brief of mine pulls from four different sources. my Gmail, my calendar, my team's ClickUp account for managing tasks and projects and just communications and all the notes in my idea verse. And because I've set it up as a recurring task inside of
Speaker A
Claude, it runs automatically every morning at 6:00 a.m. So, by the time that I'm in the kitchen, it's already waiting for me when I open my phone.
Speaker A
Now, the key to make all this work is the note you see on screen. It's the skill map. And the way to organize all these skills are to put them into systems. All these systems that you see on screen that I'm sharing now are ones
Speaker A
I actually use that I've actually built and that we have included in the linking your AI product. You can learn more about that in the description. But let's keep moving forward. So, as I scroll through my skill map, we have skills
Speaker A
that are grouped by system. And in that linking your AI product, we have an autobuilder system that actually sets up these three core files for you. And then we get to the daily trident system. So this is about managing your day. And to
Speaker A
make it work, we have two skills. The daily brief and the daily log. I just talked about the daily brief, but the daily log skill is also happening on a schedule throughout the day. It's like an interstitial journaler on your
Speaker A
behalf. That's great for both you to check in. It's like, what did I accomplish today? It's even better for AI to then bring that forward into the next day's daily brief. Next up, the Sherpa system. This one helps me map out
Speaker A
a new topic of interest so that I can learn that topic way faster. The weekly review system helps me review my week, not as a replacement, but as an additional lens. It's really helpful to make sure that nothing is slipping
Speaker A
through the cracks. I don't look at that until I've actually done my weekly review. So, I make sure that I'm always in control myself, that my thinking is leading the way, that my hand stays on the wheel. Oh boy, the rock tumbler
Speaker A
system. This gets me fast feedback on my work, not by telling me the answers or telling me how I should write something, but by asking open-ended questions. It's built on the IDI framework, imagine, discern, integrate, and it's also built
Speaker A
on K. Anders Ericson's fast feedback principle of excellence. Yes, there is a way that we can interact with AI even on semi-creative matters without losing control, without losing our voice, without abdicating what makes us unique.
Speaker A
There are also a lot of incorrect ways to do it. the chronicler system. It does the thing that so many of us wanted in the first place to save ideas and conversations before we lose them and so that we can confidently reference them
Speaker A
in the future. You're having a great conversation with AI. All you need to say is save this verbatim and it knows exactly what to do. It's going to save that conversation and it's going to place it in this history subfolder.
Speaker A
Boom. Just like that. You don't have to do a thing. Extra credit points. If you want it to be summarized, just run the summarizer skill. So then you'll have the transcript on the bottom and then the summary on top. What was the quick
Speaker A
summary? What are the takeaways? What topics were covered? Where are the next steps? All of those things in a single note. Fantastic for meetings. The janitor system. It's the thing that most people talking about AI never care to
Speaker A
mention. Actually maintaining the system. Hey, anyone can show you a few parlor tricks, but are they actually doing that cool thing? Is it really practical? Is it really reliable? Having a janitor system is how you ensure that your AI operating system can self-men
Speaker A
itself over time. And then the last one that I'd like to show you today is the courier system. This is how I share notes and ideas from my personal vault to my team simply by saying to my AI
Speaker A
assistant, share this with the team. And it will sanitize it from any personal information in that note. It will include summaries of any notes that are linked to but maybe not included in the note itself. and it will put it exactly
Speaker A
in the right place for my team. I love these and I I'm trying to give you all the details I can. We have a link below that will share even more with these skills that I use and how you can use
Speaker A
them as well. Another big question is how much obsidian do you need to know to set this up? Do you need to use the ACE folder framework or have a fully built out idea verse? The answer is no to
Speaker A
both. You can start with whatever notes you already have. The easiest path is to get access to our ultimate guide which is in the description below or available by scanning the QR code here. We'll walk you through the basics to the more
Speaker A
advanced stuff. How to set up your map files if you want to, how to set up a and more. And it will catch you up on all the Obsidian that you need to know.
Speaker A
Check that out because everything that you need to build your own AIOS is right below. We're really trying to set you up for success. Or if you missed the last few videos in this series, feel free to catch up by watching them over here.
Topics:ObsidianClaude Co-workAI operating systemNick Milonote linkingmarkdown notespersonal knowledge managementAI workflowdata privacyfutureproof AI
