YT YouTube2Text

YouTube2Text → Transcripts → Obsidian in 24 Minutes

Obsidian in 24 Minutes — Transcript

by Tina Huang · 5,469 words · 777 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I learned Obsidian for you. So, here's a
0:02cliff notes version to save you the
0:04hours and hours that I have spent
0:06building with Obsidian and AI tools. You
0:08see, Obsidian has been a very popular
0:10note-taking tool since it launched in
0:112020, but it really skyrocketed in
0:14popularity in recent years because the
0:16way that it's organized, the way that
0:18it's built is just so perfect for AI
0:20integrations. You can use it to make an
0:22AI second brain, AI databases, LLM
0:24wikis. So, in this video, we will first
0:27speedrun the Obsidian app with all of
0:29its major features. Then we'll talk
0:30about and I will demo three levels of AI
0:33integrations. And pay attention because
0:35at the end of this video, there will be
0:36a little quiz that will help you retain
0:38all of this information. Now, without
0:40further ado, let's go. A portion of this
0:42video is sponsored by Jan Spark. All
0:44right, let's start off with the
0:45fundamentals of Obsidian as a
0:47note-taking app. At its very core, it's
0:49all about just a single local folder,
0:51like this one for example. If I click
0:53into it, there are a lot of different
0:54folders and different files. And I will
0:57open that folder in Obsidian. And there
1:00you have it. This is called an Obsidian
1:01vault that contains everything in that
1:04folder. We can also open up one of those
1:05files and you can literally look at
1:07them. Amazing. Now, I'm going to
1:08speedrun all of the major features of
1:10Obsidian. And as I'm going through
1:12these, see if you can guess the three
1:13major features that make Obsidian so
1:15perfect for AI.
1:19>> [music]
1:19>> Starting off with number one. If you
1:21open up one of these files, you can see
1:22that it's really easy to read. The
1:24format of these files is what is called
1:27.md files, markdown files. It's a very
1:30standard format that's easy for you to
1:31read and pretty much any software can
1:33open it. You can also sync it with third
1:34parties if you want, like Dropbox,
1:36Google Drive. They also have their own
1:38thing called Obsidian Sync, which lets
1:40you sync up your vault with any existing
1:42remote vaults. This is actually one of
1:43the only paid features of Obsidian.
1:45Pretty much everything on it is free.
1:47And it guarantees end-to-end encryption.
1:49In fact, Obsidian is also one of the
1:51most secure apps out there. To the
1:53point, actually, in which they don't
1:54even know how many people have actually
1:56downloaded the Obsidian app? Although
1:58their CEO estimates about 5 to 10
2:00million. Okay, now let's actually do
2:02something. Let's create a new note. We
2:04can call it stuff about my life. And let
2:07me type some stuff. My name is Tina
2:10Huang and I am an entrepreneur, content
2:15creator, and run an AI education company
2:19called Lonely Octopus. I used to be a
2:24software engineer and data scientist.
2:27But for the past 6 years have been doing
2:31my own things with an amazing team of
2:34people. Very normal stuff. Now, what
2:37originally made Obsidian very famous is
2:40the fact that you can do something like
2:42this. You can do it two brackets with
2:44Lonely Octopus for example, and you can
2:46click into it and you can actually start
2:48another note. And if you open the
2:49sidebar over here, you can see the links
2:52for these notes that we just created.
2:53And we can say that Lonely Octopus is an
2:55education company that specializes in
2:57teaching ambitious people about AI and
2:59tech. And let's also do this data
3:00scientist one. Data scientists are
3:02people who analyze data and build
3:05machine learning models for a living.
3:07Great. This is what originally made
3:08Obsidian really popular. The fact that
3:10you can link notes to each other. And
3:12you can actually visualize this if you
3:14do command P or control P on a Windows
3:16computer, you can click open local
3:18graph. Let's hide this. We can see here
3:20that the stuff about my life is linked
3:22to the notes called a Lonely Octopus and
3:24data scientist. And what's really cool
3:26is that you can even do this. I have
3:28created what's called a stub. So a note
3:32that I have put here which I haven't
3:34actually created yet. Pretty cool,
3:35right? You can see that the notes that
3:37have been created are in this dark gray
3:40and the one that hasn't been created yet
3:41is in this light gray. Of course, this
3:43is only just this local graph view. You
3:46can also open up this full graph view
3:49which is of all of the different notes
3:52in my Obsidian vault. You click the
3:54anime button and it will show you how
3:56everything is being created. Very
3:58beautiful. You can also click the filter
3:59button. There's a lot of ways that you
4:00can filter, search for different things,
4:02a lot of ways that you can do displays
4:04as well, grouping things together.
4:05Honestly, mine is like a very vanilla
4:08version of this, but some people do like
4:09very aesthetic graph views of their
4:12Obsidian vaults. Obsidian also has a
4:13great search functionality. You can
4:15match by path file, tag, lot of
4:17different things. For example, if I want
4:19to look at when I reference the word
4:21salmon, I can see that oh, this is
4:23probably in reference to my food
4:25tracking.
4:26>> [laughter]
4:26>> Yes, my food tracking. Obsidian also has
4:29a lot of different community plugins
4:31that allows you to integrate it with
4:32loads of different third-party tools.
4:34There's also a lot of different custom
4:35themes, too, if you want to make your
4:37vault be very pretty. Another very
4:39attractive feature of Obsidian is the
4:41ability to embed pretty much anything
4:43that you want. You can embed images,
4:45JPEGs, PNGs, PDFs, audio recordings,
4:48>> related to AI video generation and
4:49>> tables, tweets, code blocks, pretty much
4:51anything you can think of. It's also
4:52very flexible when it comes to meta
4:55formatting and metadata. You can add
4:57tons of different types of metadata and
4:58transform them in all sorts of different
5:00ways. You can even transform your notes
5:02into different formats like tables,
5:05Kanban boards, dashboards. And finally,
5:06Obsidian has a lot of different hotkeys
5:09and shortcuts. If you do command P, it
5:11would open up a palette and you can see
5:13that there are tons of different
5:14shortcuts that allows you to zoom
5:16through Obsidian with your note-taking
5:18needs. All right, those are just some of
5:19the basics of Obsidian. There is so much
5:22more to this app and I will leave you to
5:24figure that out if you're interested in
5:26the note-taking aspects. However, right
5:28now, I have a confession to make. Come a
5:30little bit closer. You see, I actually
5:32personally rarely take notes. I am not a
5:35note-taking kind of person. I only take
5:38notes when I'm learning something and
5:39when I do that, I actually just do it
5:41with a massive Google Doc and I just
5:43like chuck things into the Google Doc. I
5:44know, I know, how disorganized and
5:46un-aesthetic of me, alas. But, I do know
5:49some people who are really into taking
5:50notes, and they are really into taking
5:52notes through Obsidian, like extensive
5:53notes and their diaries about their day
5:55and all the things that they do. Let me
5:56know in the comments where you are in
5:58the note-taking scale. Are you like 1 to
6:0010? One being like never take notes
6:02about anything, or 10 being meticulously
6:04taking notes about everything. I would
6:05be like a three, probably. But, so you
6:08may be asking now, "Tina, so why are you
6:10into Obsidian?" Well, the reason why I
6:12got into Obsidian is actually because of
6:14how compatible it is with AI. So, I want
6:17to show you guys that now. Before we go
6:18on though, I want to ask, did you catch
6:20the three major features of Obsidian? Of
6:22all the features that I just covered
6:23that make Obsidian so compatible with
6:25AI. Write down your guess in the
6:26comments, and you will get a cookie.
6:29Virtual cookie.
6:31Okay, let's move on now to Obsidian with
6:34AI. And there are three levels of this.
6:39Level one is just pointing your AI at
6:42Obsidian. Literally what its name
6:43suggests. You have your Obsidian vault
6:45that holds your notes and data about
6:47whatever topic it is that you have, and
6:49you just point an AI [music] to the
6:51Obsidian, mush them together, and voila,
6:53you get what is called an AI second
6:55brain. And this is an example of what it
6:56looks like. This is my AI second brain.
6:59It contains all types of different
7:00notes. A lot of it is related to how to
7:03make YouTube videos the way I make them.
7:06All stored in Obsidian as these markdown
7:08files. Now, to access this, I just gave
7:10Claude Co-work access to the second
7:12brain, and then I can ask it questions
7:13like, "How do I come up with a video
7:17concept?" Co-work queries it, and then
7:19it says, "I read the whole vault before
7:21answering." Says, "Your notes say that
7:23you don't have a concept generation
7:24problem." Great, thank you. And it says,
7:26"The process your notes already describe
7:27includes coming up with a candidate,
7:29filtering it by six different criteria
7:31in this file that is listed in this note
7:34over here, and then it got to build it
7:36based upon this note over here." Great.
7:38And then it just goes on to talk about
7:40more stuff, and then I can like have
7:42full-on conversations with the notes
7:43that I have. You, of course, are not
7:45just limited to using Cohere. When I
7:47want to build something based upon my
7:49notes, I actually like using Codex. I
7:51can ask Codex, "Build me a checklist of
7:55all these signals to determine if a
7:59video concept is worth making or not."
8:03It would also read through my entire
8:04second brain, and it would eventually
8:06build me something like this checklist:
8:08Is this worth filming? And I can check
8:10things off to ultimately decide if I
8:13want to build a video or not. Amazing.
8:15[music] Great. So, how does this all
8:16work under the hood? Let me go into more
8:18details. So, in a traditional second
8:20brain, which is more recently
8:21popularized by Tiago Forte, is that your
8:23brain is good at thinking, not for
8:25storing information. So, you are meant
8:27to write down all the information that
8:29you have in your brain, and then
8:30retrieve it when you want to. Like, for
8:31example, maybe when you met Bobby, Bobby
8:34told you his favorite color was yellow.
8:35So, you wrote, "Bobby's favorite color
8:37is yellow," and you put that into your
8:38second brain, your Obsidian vault. And
8:40one day, maybe you want to buy a gift
8:41for Bobby, and you would you maybe you
8:44like search through your second brain,
8:45and you what's Bobby's favorite color,
8:47and you find out that it's yellow.
8:48You're like, "Ah, yes, yellow," and you
8:49can go buy a gift, right? That's the
8:50concept. Sorry, it's a little trivial.
8:52That is like the concept of having a
8:55traditional second brain. Now, in AI
8:56second brain, you're still writing these
8:58notes. However, you have the AI go and
9:01retrieve it for you. Not only can it
9:03retrieve it, it can also interpret it
9:04and analyze the information, too. And
9:06because of the way Obsidian's
9:07structured, it's basically just a bunch
9:08of markdown files, right, in this local
9:11folder. You can point the AI at it, and
9:12it's able to understand this information
9:14and analyze it and interpret it and
9:16answer the questions that you want. You
9:17can actually use all sorts of different
9:18types of AI. For me, I like to go with
9:20Claude Cohere. It is a local AI agent,
9:23Anthropic's version, that's like very
9:24simple and has a nice UI. But sometimes
9:26I would also use Codex if I want to use
9:28the data programmatically. For example,
9:29if I want to do something more like
9:30analysis on it, build like an app, a
9:32dashboard, something like that, I would
9:34use a coding agent like Codex or Claude
9:36Code. Or if I wanted to be super duper
9:38completely private, I would use an open
9:40source agent harness like Hermes with a
9:42local model or deep seek harness or open
9:45code. Not going to go into much more
9:46detail about how these harnesses work
9:47and how local AI agents work. If you
9:49want more details about this, please do
9:50check out this video over here which I
9:52go into more details. Anyways, another
9:54way that you can connect your AI is
9:56through the extensive network of
9:57Obsidian community plugins. Stuff like
9:59smart connections and co-pilot. They can
10:01plug directly into your Obsidian vault
10:03and be able to do stuff and answer
10:04questions for you. So, like I said
10:06earlier, I am not a note taker kind of
10:08person, especially not about like my
10:09personal life. I guess I prefer to live
10:11each day in a way in which I don't
10:13remember what happened the day before or
10:15the week before. But for those of you
10:17who are note takers, I would have to say
10:19I am really really jealous of you cuz
10:21you probably have a bunch of notes, this
10:22whole corpus of notes about so many
10:24different types of things from years and
10:26years and years. And I think you can
10:27probably learn some really cool insights
10:29about your life, yourself, other people,
10:32whatever you take notes on, by putting
10:33your notes into an Obsidian vault and
10:35connecting your choice of AI to it. Not
10:37going to go into too much more
10:38implementation detail about this. But I
10:41will put on screen now a summary slide
10:42including specs of how to build an AI
10:45second brain. You can take a screenshot.
10:46Or you can check it out in the free
10:48resource that I have linked in the
10:49description, just like my other videos.
10:51In the resource, I also link some
10:52additional resources. If you do want to
10:54implement the AI second brain. Big shout
10:56out to Tiago Forte Forte and Nick Milo
10:59who I learned a lot from about second
11:02brains in general. Now, all these notes
11:03and data that I have in Obsidian
11:05probably make me look like I am a very
11:06in control person. But what I definitely
11:08do not have in control is my raw email
11:11inbox. I am low-key scared of my inbox.
11:13Like right now, I have 17,000 emails in
11:16my work inbox. I get hundreds of
11:18inquiries every day. All of this buried
11:19under newsletters and things that I
11:21apparently signed up for and I don't
11:22remember. But now I'm using Genmail by
11:24GenSpark to handle all of this. It's an
11:26email agent that handles your inbox like
11:2810 times faster and it learns how you
11:29write. Genmail sorts everything by
11:31priority, so I only see what actually
11:33needs my attention. Like the brand reach
11:34out from yesterday and the invoices that
11:36I'm handling. These are on top, while
11:37everything else is nicely tucked away.
11:39What used to take me like an hour of
11:40scrolling only takes me 5 minutes now.
11:42Jan mail also learns from my past
11:43replies, my tone, the way that I talk to
11:45different people. Every draft that I
11:47generate increasingly sounds like me,
11:48not just a template. Of course, I still
11:50always do the final review, maybe tweak
11:51things a little bit, and then just press
11:53send. I also set up one rule. Any
11:54invoice email gets auto forwarded to my
11:56finance folder. This way I don't have to
11:58think about it again. This is so small
12:00me so much time and money during tax
12:02season. And because Jan mail is part of
12:03the entire Jen Spark ecosystem, I can
12:05take a long email thread and turn it
12:07into a doc or a slide outline without
12:10leaving the app. You can download Jan
12:11mail for free in the link in the
12:12description. Thank you so much Jen Spark
12:13for sponsoring this portion of the
12:14video. Now, back to the video. Moving on
12:16to level two, which I call the AI
12:18database.
12:22You see, for the AI second brain, when
12:24you point AI at your city vault, you are
12:26the one that writes the notes, while AI
12:28is the one that retrieves these notes
12:30and interprets it, analyzes and answers
12:31your questions from these notes. [music]
12:32For level two, it's the other way
12:34around. This is about AI writing notes
12:37into your Obsidian vault, which is
12:39actually much more suitable for people
12:40like me who don't like writing notes.
12:42Our company documentation system and my
12:44personal productivity system is based
12:46upon this structure. Let me show you
12:47what it looks like. This is my AI
12:50database. Woah, see, it has so much data
12:53in it. Life Bot, who is my AI
12:54productivity and health coach, writes
12:56stuff to it. For example, I can give it
12:58this picture and say, "Log this for me."
13:02It's just this picture of a green tea
13:03that I just had. Trying to log more of
13:05my food and drinks, and it's able to log
13:07it for me. It's an unsweetened green tea
13:10with zero calories. Trying to do [music]
13:12OMAD, by the way. Let me know in the
13:14comments if anybody else is doing OMAD.
13:15There is also Taco Bot, who is my AI
13:17assistant and COO, and I can say,
13:19"Create
13:20doc on how YouTube topics engine works."
13:24It is doing its thing, and it wrote the
13:26process documentation to the Obsidian
13:29vault. I also have these little desktop
13:31widgets that write these Pomodoro
13:33[music] logs and my to-do lists all to
13:35this Obsidian vault. And there are a lot
13:37of other things I write [music] to this
13:39vault as well. So, you may be wondering
13:41though, what is the purpose of having
13:43all this data in this vault? Well, I
13:44guess one thing is just nice to have
13:46these notes that are here, like logs
13:47about my life and the company, cuz I'm
13:49not going to take these notes myself.
13:51But, on top of that, I do give access to
13:53my AI agents like my spot here. And
13:55because it has access to this AI
13:57database, I can ask it questions like
13:59how do I improve my productivity, deep
14:02work sessions based on my personal data.
14:06And we can see that it's accessing this
14:08data, and it can give me very specific
14:10tips like protecting when it is that I
14:12should be working because it has data
14:13about when I've been working, how to
14:15sequence what I'm working on, like
14:16knowing that scripting is my hardest
14:18task and doing tasks where operations is
14:21my easiest task, when to take breaks,
14:23etc., etc. This is so incredibly
14:25powerful. I have an entire other video
14:26explaining my entire productivity
14:28system. If you want more details about
14:29that. Same story when it comes to our
14:31work dynamics and our work productivity,
14:33too. Having this data has allowed the
14:35team in general to be much more
14:37productive, as [music] well. So, how
14:38does this work under the hood? Well, the
14:40concept is that there are a lot of AI
14:41agents and AI systems that are all
14:43writing into this Obsidian vault. You
14:45can think about this Obsidian vault as
14:47becoming this database where all these
14:49AI tools are logging information into. I
14:51also do have the Obsidian sync, if you
14:53remember what that is from earlier,
14:55which allows me to have machines be able
14:57to simultaneously access the Obsidian
14:59vault at the same time. And it's all
15:01like end-to-end encrypted. So, I have AI
15:02agents running on my personal computer
15:04that's writing to the vault. There are
15:05AI agents on my Mac Studio that's
15:07writing into the vault. And there's also
15:09AI agents on my VPS that is writing to
15:10the vault, as well. And the vault is
15:12synced across these different machines.
15:14My favorite AI models that I use to do
15:16all of this writing is Gemini Flash,
15:18Deep Seek V4 Pro. And if I want a
15:20completely local private option, I use
15:22the Qwen 3.6 36B model that runs locally
15:25on my Mac Studio. That is like super
15:27duper complete privacy. I also track my
15:29health data originally, which is using
15:30Apple Health, but I actually recently
15:32got this Aura ring. So, I'm literally
15:34just trying out with it. This is like my
15:36third day of using it. So, hopefully it
15:37gives me even more health information.
15:39Thank you to everybody who suggested
15:40this to me. I was debating between the
15:42Aura ring, Apple Watch, and Whoop, and
15:44ultimately went with the Aura ring cuz
15:46some of you guys said that it was very
15:47good. Too early for me to tell right
15:49now, but I will let you guys know.
15:50Anyways, so yes, I have all these AI
15:51agents, all these different softwares
15:53and tools writing into this Obsidian
15:55vault to create this database. So, I can
15:57read this through like with my human
15:58eyes if I want and pretty much do
16:00whatever I want with it, edit it,
16:01anything that I like. I personally also
16:03like connecting my Hermes agents into
16:04this database, so I'm able to like query
16:06the database and get a lot of
16:07information. So, there you have it. AI
16:09writing to a database and you doing
16:10whatever you want with the notes. I will
16:12put a summary slide and specs of my AI
16:15database on screen right now if you want
16:17to check it out right now. Would
16:19recommend trying out this level first if
16:21you are someone who's interested in
16:22gathering like a rich database of data
16:25and notes. Finally, let's move on to
16:27level three, the LLM Wiki.
16:32So, level three is the full complete AI
16:34managed infrastructure turned by Andre
16:36Karpathy, who is like a superstar AI
16:39thought leader and engineer as an LLM
16:42Wiki. This is when you have AI write the
16:44notes for you, manage all of it for you,
16:46and retrieve it, analyze it, interpret
16:48it for you, too. You literally don't
16:49even touch your Obsidian vault. As Andre
16:51Karpathy puts it, Obsidian is treated as
16:53the IDE, the LLM acts as the programmer,
16:56and the vault itself is the code base.
16:58You're just like a observer. Let me show
17:01you a demo of this with my Hermes LLM
17:03Wiki, which I have been slowly building
17:05and using as I deep dive into Hermes cuz
17:07I'm like really into building my Hermes
17:09out. This is my LLM Wiki about Hermes.
17:12It has different notes and resources
17:14that I've collected in the past few
17:16months. For example, I come across this
17:18really cool resource that talks about
17:20how to do a Hermes Kanban board. So, I
17:22can copy this and give it to my Wiki
17:24Bot, which is a Hermes AI agent. It
17:27ingests and processes information and
17:29saves it. Here it is. I can also ask the
17:32LM Wiki questions like, "How do I use a
17:35Hermes Kanban board?" And it would read
17:37through my Wiki and then give me a
17:39step-by-step based upon what is stored
17:41inside the Wiki. It has an index of
17:43everything that's contained inside the
17:44Wiki and keeps a log of everything that
17:46it does.
17:47>> [music]
17:47>> Periodically, you can also ask it to
17:48please do a health check. And it would
17:51go in there and just check through
17:53everything, running any stale notes, and
17:55deal with any contradictions that are
17:56there, too. Honestly, so cool. It's
17:59completely managed by an AI. So, this
18:02seems pretty magical, right? It
18:03literally just operates itself. But, you
18:05can actually set up something like this
18:06in less than 5 minutes, which I will
18:08tell you how in just a little bit. But,
18:09let's actually first take a look under
18:11the hood of how this works. So, this is
18:14the original LM Wiki gist that Karpathy
18:17wrote. It says it's a pattern for
18:18building personal knowledge bases using
18:20LMs. The core idea is that the LM
18:22incrementally builds and maintains a
18:24persistent Wiki. You never or rarely
18:26write the Wiki yourself. The LM writes
18:28and maintains all of it. You're in
18:30charge of sourcing, exploration, and
18:31asking the right questions. The LM does
18:33all the grunt work to summarizing,
18:35cross-referencing, filing, and
18:36bookkeeping that makes a knowledge base
18:37actually useful over time. In practice,
18:39the way that he does it is that I have
18:40the LM agent open on one side and
18:43Obsidian open on the other, where he
18:44stores all of his notes. Blah, blah,
18:46blah. Okay, and he also says that this
18:47can apply to a lot of different
18:48contexts. For example, personal is like
18:51for tracking your own goals, health,
18:52psychology, self-improvement, research,
18:54going deep on a topic over weeks or
18:55months is what I did with my Hermes LM
18:57Wiki, reading a book, filling out
18:59chapter as you go, building up pages for
19:01characters, themes, plot threads, and
19:03how they connect. Business and teams,
19:05having internal Wiki maintained by LMs
19:07fed by Slack threads, meeting
19:09transcripts, project documentations, and
19:10customer calls. I am currently exploring
19:12this option taking our team
19:14documentation management, which is at
19:16level two right now, having the AI
19:17database and bring it up to level three,
19:19having it as an LLM wiki. And another
19:21option is having competitive analysis,
19:22due diligence, trip planning, course
19:24notes, and hobby deep dives. Anything
19:25basically where we're accumulating
19:27knowledge over time and want it
19:28organized rather than scattered. So,
19:29here is the three layers of it. You have
19:31your raw sources, which is your curated
19:33collection of source documents, your
19:35articles, papers, images, and data
19:37files. So, these are immutable. The LLM
19:39reads these but never actually modifies
19:40them. This is your source of truth, your
19:42notes. Then you have the wiki, which is
19:44a directory of LLM generated markdown
19:46files, summaries, entity pages, concept
19:48pages, comparisons, and overview, a
19:50synthesis. So, the LLM owns this layer
19:53completely. It creates these pages and
19:55documents it and updates it as new
19:56sources arrive. Then you have the
19:58schema, which is a document like a
20:00cloud.md for cloud code or agents.md for
20:03codex that tells the LLM how the wiki is
20:05structured. So, it explains how it's
20:07able to read the corpus. Stay with me
20:08here, okay? I promise you that it's
20:10simpler than it seems. So, you basically
20:11just have these three layers. [music]
20:13Your raw sources is the only layer in
20:15which the LLM doesn't touch. These are
20:17just like everything that you put into
20:19it. Like in my demo, it's all of these
20:21articles that I found in documentations
20:23and videos, as well as my own notes
20:25about Hermes. Then he explains there are
20:27three operations that your AI does in
20:30your LLM wiki. The first one is
20:32ingesting. When you drop a new source
20:34into the raw collection and tell the LLM
20:35to process it. So, in my case, I use
20:37Telegram as my kind of port in which I
20:40would tell my AI, in this case it was a
20:42Hermes agent, to process the sources
20:44that I would give it and then it would
20:45store this information. Then the next
20:47one, the next operation is querying.
20:48This is when you want to get information
20:49from your wiki. So, you can ask
20:50questions against the wiki in which the
20:52LLM will search for relevant pages and
20:54information, reads them, synthesizes
20:55them, and tells you stuff. Like when I
20:56ask my wiki, "How do I build a [music]
20:59Kanban dashboard?" It's able to query
21:01the documentations and all the articles
21:02that I have about Kanban dashboards for
21:04Hermes and it gives me information about
21:05how I can implement this. And finally,
21:07it has the lint operation. This is when
21:09the LLM health checks the wiki. It looks
21:11for contradictions between pages, stale
21:13claims that newer sources have
21:14superseded, orphan pages, no inbound
21:16links, basically cleaning it up and
21:17maintaining it. It's like pruning the
21:19wiki. This keeps the wiki healthy as it
21:21grows. And finally, there are two
21:23special files that helps the LLM and you
21:26to navigate the wiki as it grows. The
21:28first one is an index.md file. The MD is
21:30a markdown file is content-oriented. So,
21:32this is the catalog of everything in the
21:34wiki. For example, this is for my Hermes
21:37wiki. Each page is listed and it has
21:39information about it with a link,
21:41one-line summary, and even like metadata
21:42like date or source counts. And things
21:44are organized by categories. The LLM
21:46updates this [music] by itself on every
21:48ingest of additional sources. This is
21:51what helps your AI, your LLM, be able to
21:53navigate your wiki to figure out where
21:55is the right places to look to get the
21:56information that you want. This becomes
21:58increasingly more important as your wiki
22:00gets bigger and bigger. And then
22:01finally, there is the log.md, which is
22:04chronological. This is the one that I
22:05have for my Hermes wiki. It is literally
22:07an append-only record of everything that
22:10happens in the wiki. What sources are
22:11being ingested, what queries are being
22:13done, lint passes. This is just a log to
22:15be able to document everything that is
22:18happening within the wiki. It gives you
22:20a timeline of the wiki's evolution and
22:22is also a way for you to know what's
22:24happening. And if something goes wrong,
22:25it's also a way for you to go in there
22:26and to figure out what happened. And
22:28literally, [music] that's it. There are
22:29like, you know, a bunch of tips and
22:30tricks that he suggests and, you know,
22:32more information and things like that
22:33and people like discuss a lot about it
22:34as well. Which you can dive deeper into
22:36later. But basically, that is like the
22:37full concept of it. And remember what I
22:39told you earlier, how this may sound
22:41like a lot, but it's actually really
22:43easy to implement these days. Well, that
22:45is because so many people have actually
22:47implemented this. For me, I use Hermes
22:49to do this and there is literally a
22:51skill called LLM wiki. All I did was I
22:53created a new profile, a new Hermes
22:55agent, I invoked the LLM wiki skill, and
22:57I attached a Telegram bot, which is how
22:59I I along these sources to my wiki. And
23:01that's it. And that's how I started
23:03building my wiki about Hermes and a
23:05bunch of other topics, too. You also
23:06don't need to use Hermes for this. You
23:07can use all sorts of different types of
23:09AIs. The best documented is actually
23:11with Claude code. There are so many
23:13GitHubs that you can just get clone,
23:14follow the steps, and have a wiki
23:16running, and start doing stuff with it
23:17using Claude code. You can also use like
23:19Codex, open code, pretty much like any
23:21type of AI. The whole point, as Karpathy
23:24states, is that this is actually kept
23:25generally vague. His description of it
23:27is generally vague. So, you could
23:28literally just like copy-paste this and
23:31give it to any AI and tell it to
23:32implement it, and it should be able to
23:34come up with some way of doing this.
23:35This is a pattern that can be
23:37implemented with any type of AI.
23:39Actually, any type of database, too.
23:40But, you know, this is an Obsidian
23:41video, so that's how we implemented
23:43using Obsidian. And honestly, Obsidian
23:45is the go-to for making these LM wikis.
23:47Even Karpathy himself uses Obsidian.
23:50Because Obsidian is so good for
23:51integrating with AI. All right, I'm
23:52going to put now the summary slide on
23:54screen with the specs for the LM wiki
23:56and implementation details. Please take
23:58a screenshot or check out the free guide
24:01that I put in the description, which I
24:02will also add in additional resources
24:04that you can check out if you're
24:05interested in implementing this. And
24:07that's it. Yay! Thank you so much for
24:08watching until the end of this video. As
24:10promised, here is a little quiz. Please
24:12answer these questions in the comments.
24:13Because research shows that immediately
24:15reviewing information is the best way to
24:17retain that information. And you don't
24:19want all of that information to be going
24:20to waste, do you? Especially since we're
24:21talking about things like second brains
24:23and memory databases and storing
24:25information. In the end, your brain is
24:27the smartest. Thank you so much for
24:28watching. I hope this is helpful for you
24:30and inspires you to try Obsidian, do
24:32more stuff with it, and I will see you
24:34guys in the next video or livestream.

About this transcript

This page contains the full transcript of Obsidian in 24 Minutes by Tina Huang, generated from the public captions YouTube serves with the video. The transcript has 5,469 words across 777 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.