Como construir um segundo cérebro (guia definitivo) — Transcript
Full transcript
- 0:01Every day a new tool comes out, a new
- 0:02LLM is released, something that makes
- 0:04you think: "Darn, I'm falling behind."
- 0:07But there is one single way you can
- 0:08shield yourself from this. And that way
- 0:11is what everyone, in all my YouTube
- 0:13content, is always asking for, and it's
- 0:16called a second brain.
- 0:18And "second brain" is just a fancy name
- 0:20for nothing more than a folder with an
- 0:22organization system and files with some
- 0:24tools attached that boost your second
- 0:27brain.
- 0:28And in this video, I want to bring you
- 0:30the concept from ground zero, meaning,
- 0:32explain to you what a second brain is
- 0:34and go through three steps for you to go
- 0:37from basic, to intermediate, and reach
- 0:39advanced.
- 0:40So this is a video where I will explain
- 0:42to you how you can build a second brain
- 0:44, give your second brain a nice upgrade
- 0:46, and how you can organize and architect
- 0:48it so you have flexibility, mobility, so
- 0:51you aren't held hostage by any company,
- 0:53and so you never lose context.
- 0:55In other words, your brain will never
- 0:57forget anything and you can use any
- 0:59company or any AI tool available from
- 1:01here on out. Let's get to today's video
- 1:03. If you are new here to the channel, I
- 1:04am Bruno Camoto and it is a giant
- 1:05pleasure to have you here with me. I
- 1:06spent the last 10 years of my life
- 1:08building startups and, suddenly, I left
- 1:10that startup market. I entered the SaaS
- 1:12world, spent 4 years talking about
- 1:14micro-SaaS, built the largest micro-SaaS
- 1:16community in Brazil, and today, I
- 1:17dedicate my time and energy to building
- 1:19Pixel AI, which is my AI agent
- 1:21implementation company for businesses.
- 1:23So, man, I love this topic, I study it a
- 1:25lot, I live off this, I have over 1,000
- 1:27clients that I serve and take care of.
- 1:29So, man, it's a topic that I understand
- 1:31reasonably well enough to record this
- 1:33video and come talk to you guys. And
- 1:35today I want to talk about the second
- 1:37brain, which in my view is the most
- 1:38important thing of all in 2026.
- 1:40So, if you don't know what a second
- 1:42brain is, in a very simplistic way of
- 1:43putting it, the second brain is nothing
- 1:45more than a folder.
- 1:47It means it's a series of text files,
- 1:49right?
- 1:50So, the vast majority of text files are
- 1:52markdown, and it's a folder where you
- 1:54can simply organize things. So, in a
- 1:56very raw way of putting it, the second
- 1:57brain today, which is this fancy name,
- 1:59it tends to have several appearances,
- 2:00right? So, people on the internet go
- 2:03there and put up a second brain full of
- 2:06constellations, showing how various
- 2:08little dots clump together and how
- 2:11everything connects, right? And it looks
- 2:13very pretty. And I could open up 20
- 2:15examples for you, but I won't waste time
- 2:16with that. On the other hand, I will
- 2:18spend 5 seconds here for you. So, this
- 2:20right here is an example of a second
- 2:22brain. And this second brain has several
- 2:24forms, doesn't it?
- 2:26From more complex ways to do it with
- 2:28tags and a more technical structure, to
- 2:31a very simple structure, which is
- 2:33nothing more than a second brain based
- 2:36on a folder system.
- 2:38And a second brain based on a folder
- 2:40system is nothing more than, well,
- 2:41folders, right?
- 2:43So here you have one folder, here you
- 2:45have two folders, here you have three,
- 2:46here you have four, here you have five
- 2:48folders, six, seven, blah blah blah.
- 2:50So, in a second brain, you can have
- 2:51several folders.
- 2:53You can have a folder called areas of
- 2:55your company, one called skills; you can
- 2:57have a folder about projects, one we can
- 3:00have about content, one we can have
- 3:02about materials, another we can have
- 3:04about finances, and then, I don't know,
- 3:06you can create whatever folders you want
- 3:08.
- 3:09These folders can also be right here on
- 3:11your computer. So, here I have several
- 3:14folders for several projects, and this
- 3:16second brain folder here is my official
- 3:18folder, right?
- 3:19So here I have the whole structure of my
- 3:21Bruno Okamoto second brain, which,
- 3:23clearly, lacking creativity, I called
- 3:25Okamoto Second Brain. So this second
- 3:27folder has the whole structure, right,
- 3:29from how it operates to all my content
- 3:31and context. And this, man, we are
- 3:33talking about context captured over a
- 3:34long time, it's almost a year of context
- 3:36that I have here, inside these folders,
- 3:38right? So I have personal stuff, I have
- 3:40marketing, I have my companies, I have
- 3:42branding, I have, anyway, a ton of
- 3:44really cool things in here, and it's
- 3:46nothing more than my second brain. And
- 3:48as you can also see, this second brain
- 3:50can have several structures. The vast
- 3:52majority are just text structures, right
- 3:54?
- 3:55So we have here my agents, there's even
- 3:57the way Claude operates inside this
- 3:59structure of mine, because my brain is
- 4:01portable.
- 4:03So I open it in Claude, I open it in
- 4:04Hermes, I open it in Grokbot, which I
- 4:06also subscribe to, I open it anywhere,
- 4:08right?
- 4:09So it is very flexible. So now that you
- 4:11understand the concept of a second brain
- 4:13, you can see that it's nothing more
- 4:14than a sequence of folders. So let's
- 4:16talk about how you can build a second
- 4:18brain efficiently, effectively, and stay
- 4:20mobile, stay fast, never lose context,
- 4:22and never forget. There are three major
- 4:24steps that we are going to talk about in
- 4:26this video. The first step is a free
- 4:27step, so you don't need to spend any
- 4:29money on it and it's very simple. So the
- 4:32first step is for you to work in a free
- 4:34way.
- 4:35So zero cost here for you to build the
- 4:37foundation of this project, of your
- 4:39second brain.
- 4:40So the tool I recommend for everyone to
- 4:42set up a second brain is a tool called
- 4:44GitHub. Why do I recommend GitHub?
- 4:46Because GitHub is a tool made for
- 4:48developers, but it works very well for
- 4:51you to work with AI. I say it is a tool
- 4:53made for AI, so it works really well.
- 4:56Microsoft even bought this company, so
- 4:57it's a secure business, it's super safe
- 4:59, it has 2FA authentication, you can,
- 5:01man, just go for it, it will work out
- 5:03fine, okay? So GitHub is a free tool
- 5:05that you can use and put everything
- 5:08inside it. So what you have to do is
- 5:10create your second brain, which is
- 5:12nothing more than creating a folder on
- 5:14your computer.
- 5:15So here we will create a little folder,
- 5:17let's call it the brain folder.
- 5:19So inside this folder, which is
- 5:21basically just clicking on the desktop
- 5:22and creating a folder, you will start
- 5:24putting your things in there.
- 5:26So here we can put, for example, we can
- 5:29put our company, right? Here we can also
- 5:31put our personal stuff. So you can put a
- 5:34folder here and divide it into
- 5:36subfolders if you want.
- 5:38And then inside company, you put your
- 5:41projects, you put the areas where you
- 5:43have files, you put the tools you use
- 5:45inside a text file, right?
- 5:48You can put, man, Word, you can put
- 5:50Notion, Docs, it doesn't matter. You can
- 5:53plug multiple tools in here. So you keep
- 5:56giving context to the AI, which is the
- 5:57only way for the AI to understand and
- 5:59work with everything you need for your
- 6:01business and your day-to-day life.
- 6:03And then in personal you can also put
- 6:05finances here, so you can connect a
- 6:07plug-in, for example, to capture your
- 6:09bank statement.
- 6:12So it has a way to capture your bank
- 6:14statement for reading, and then you can
- 6:15automate your finances too, you're
- 6:17welcome for the tip, just as you can
- 6:19also put your personal controls, bills,
- 6:21home, etc. here, and you organize
- 6:22everything.
- 6:24So this folder is the folder where your
- 6:26AI, whether it's Claude, ChatGPT, or
- 6:28Hermes, will be able to have the basics
- 6:30to work with you.
- 6:32And from the moment you start putting
- 6:34everything in this folder, your AI
- 6:35starts to have more and more context
- 6:37about everything you're doing and can
- 6:39help you more surgically.
- 6:41So, when you reach this first step and
- 6:43have set up this basic little folder
- 6:45project, my first tip for you is to
- 6:46build these canonical files.
- 6:49So, these files here, they are very
- 6:51interesting. The map is an index that
- 6:53you ask your agent to read first.
- 6:56So, you come here to your agent and you
- 6:58put a file called map inside its folder
- 7:00.
- 7:01The map.
- 7:02MD is nothing more than a series of
- 7:04instructions that your agent will read
- 7:07to find the folders.
- 7:08So, the map. MD, for example, it will
- 7:11read what the files are, what each file
- 7:13does, what each folder contains, and so
- 7:15on. It's a small, it's a small map,
- 7:18literally. Yes, it's a small map.
- 7:21So, it's an index that allows your agent
- 7:23to locate information faster every time
- 7:24you open your folder, every time you
- 7:26start a session, and so on.
- 7:28And that's where my tips for you come in
- 7:31. My first tip is for you to create a
- 7:33file called hot.
- 7:34MD, which is what is alive. So, the hot
- 7:36file pulls this from various places,
- 7:38right? The example isn't here, but
- 7:40basically, it pulls it in.
- 7:43So from the moment you connect your
- 7:44email, your calendar, you connect your
- 7:46CRM, you connect your WhatsApp, you
- 7:48connect anything you want to connect,
- 7:50you can ask it to create a small
- 7:51automation to keep saving everything
- 7:53that is hot in your daily life.
- 7:56So, my agent, Amora, she saves every day
- 7:58from a meeting I had, from my WhatsApp
- 8:01that she read, from a YouTube video I
- 8:04recorded. It saves what is hot, which
- 8:06are things I need to do, things on my
- 8:09calendar, it's all in my Notion; the
- 8:11source doesn't matter. The second thing
- 8:13is a file called People. So, People is
- 8:16where it creates a small CRM of every
- 8:18person I have interacted with. So, if
- 8:20you sent me a message on my WhatsApp,
- 8:22you ended up in my People, right? And
- 8:24then you say to me: "But Bruno, do you
- 8:25authorize everyone to enter your People
- 8:27?" No, you can set up small rules, right
- 8:29? So, every day my agent sends a message
- 8:31on my Telegram with a list of people
- 8:33saying who I authorize to enter the
- 8:35brain. So, once I authorize that person
- 8:37, they go into my People, into my
- 8:39pending items, into my hot list; they
- 8:41enter the whole structure. Just like I
- 8:43have pending items and deadlines, right
- 8:44? So, we have pending items and
- 8:46deadlines here.
- 8:47So, if Ana sent me a message on WhatsApp
- 8:49and she is on the list of people
- 8:51approved for my second brain, what she
- 8:53sent can become hot or a pending item
- 8:56based on my agent's or my AI's digestion
- 8:58.
- 8:58"Oh, but Bruno, how do you perform this
- 9:00digestion?" It's very simple, they are
- 9:02skills. So, you first create a raw log,
- 9:05right? So your AI needs a daily routine
- 9:07to pull a log from WhatsApp, email, the
- 9:10calendar, and so on. And from that raw
- 9:13log, you will perform a digestion. And
- 9:15for this digestion, you will use a
- 9:17cheaper LLM, like Luna, or even Sol, and
- 9:20you have that LLM read and digest
- 9:22everything. And from that digestion, it
- 9:25updates the hot list, People, pending
- 9:27items, decisions, diary, and so on, okay
- 9:29? So, in here you have the sessions. So
- 9:30, for the sessions, you basically create
- 9:32a skill where you ask your AI to save
- 9:34the entire session of everything you
- 9:36talk about every day. So, Hermes already
- 9:38does this automatically by default.
- 9:40When you compress the session, it has
- 9:42what it calls a 75%threshold, so when it
- 9:44hits 75%of the session, it compacts it,
- 9:46right?
- 9:47Claudinho also has this structure here.
- 9:49So, for Claudinho, you can create a
- 9:51skill called "save," for example, that
- 9:53does exactly that. So, that's the end of
- 9:55a session that closes two brains at once
- 9:57. So it performs semantic propagation,
- 9:59right, in the Mora brain: pending items
- 10:00, decisions, people, projects, metrics,
- 10:02skills, and maps, you see? This is very
- 10:04cool. And it also pushes, right, into my
- 10:06company's brain, it's to save things
- 10:09from my work there, right? So this is
- 10:11very cool, this structure here helps a
- 10:13lot. And then you have the memory, right
- 10:15, which we already produce automatically
- 10:17, and it saves the memory here about who
- 10:20the user is and also daily notes. So
- 10:22this is step one. It's a little step
- 10:24where you have zero cost to get the
- 10:26basics of your brain. So, a quick
- 10:28conclusion here. You then have your
- 10:31brain folder and you have the canonical
- 10:33files, which are these main files, okay
- 10:36? And it's very easy for you to digest
- 10:38this by skill and so on. Take this
- 10:39video's description, send it to Soiá,
- 10:41and she'll help you set up these files,
- 10:43set up the digestion system, and
- 10:44everything else. And before moving on to
- 10:46the next step, I want to talk about our
- 10:48sponsor for today's video, which is
- 10:50Hotmart. So, Hotmart is our official
- 10:52gateway that we use here at Pixel. And
- 10:54man, honestly, I really like it,
- 10:55straight talk here, we really like it,
- 10:57especially because of the API. I
- 10:59personally use their API a lot to know
- 11:01what is happening within my courses,
- 11:03within my products, within my finances,
- 11:05and so on. I also think their
- 11:06installment system is very intelligent.
- 11:08I also really like their customer
- 11:09support. We get excellent support, I
- 11:11also really like the documentation, the
- 11:13ease of use, anyway. It's a gateway that
- 11:15in my view is like light-years ahead of
- 11:16the competition and I highly recommend
- 11:18it. Thank you Hotmart for sponsoring
- 11:20this video here. Let's go to the second
- 11:22step, which is the level. So here you
- 11:24already start spending a little bit.
- 11:27So why do you start spending a little
- 11:29bit of money? Because you have to have a
- 11:31VPS, okay? So in this case, I recommend
- 11:33Hostinger, which is the VPS I always use
- 11:35, I always recommend. I think Hostinger
- 11:37is the best VPS by far. And Hostinger
- 11:39has this 43 buck plan that you can come
- 11:41here and get. Or you can get this 29 one
- 11:44, it covers it, but man, I think it's
- 11:46too little. So maybe this 100 one works
- 11:48well. You can come here to choose your
- 11:50plan and use the coupon that's in my
- 11:51description along with the link.
- 11:53And you can enter your coupon here,
- 11:55apply it, and you'll get an extra 10%
- 11:57discount.
- 11:58Then you come here, set it to Ubuntu,
- 12:00choose the version they recommend, which
- 12:02is 24.04, confirm, and create your VPS.
- 12:05Simple as that. So, what are you going
- 12:06to do in this second step here? We are
- 12:08going to install a directory called
- 12:10JBrain.
- 12:11JBrain is when you start to index your
- 12:14files offline.
- 12:16And it starts to work in a much simpler
- 12:19way for you to find things, right?
- 12:21Finding files, finding documents,
- 12:23finding texts, finding everything. All
- 12:25right? So, JBrain works really well for
- 12:28several tools.
- 12:29It works for Grok, Botkit, Claude,
- 12:32Hermes, and OpenAI. It's kind of a
- 12:35wildcard, right? It was built by Gary
- 12:37Tan, who is the president and CEO of Y
- 12:39Combinator. And he said his challenge is
- 12:42that he has like 1,500 pages, 25,000
- 12:44people, and 5,340 companies within his
- 12:46structure.
- 12:47And he said: "Man, I need to take and
- 12:49look at all these meetings, emails,
- 12:51tweets, everything I'm doing.
- 12:53This stuff needs to be processed, it
- 12:55needs to be enriched, it needs to be
- 12:57digested, and the structure of an agent
- 12:58wasn't made to hold that much data."
- 13:01So he built JBrain in a very simple way
- 13:03, so to speak. And JBrain is nothing
- 13:05more than you taking this repository as
- 13:07well, which I'll leave the link for in
- 13:09the description, and asking your AI to
- 13:10install JBrain. It doesn't matter which
- 13:12AI you install it on. You just send the
- 13:14repository to it and say: "Install
- 13:16JBrain, test it, and tell me if it's
- 13:17working." I won't even leave the prompt
- 13:19, man, because it's ridiculously simple
- 13:21. Once you run the prompt, right? So
- 13:23let's put it here, look. Claude, I want
- 13:25you to install JBrain here, run a test,
- 13:29and show me how it works. Very simple
- 13:31like that. It will install JBrain here.
- 13:33Let's let Opus think for a little bit.
- 13:35And then we'll talk about JBrain. JBrain
- 13:38has this response with sources, okay? It
- 13:40also uses an API for indexing, but you
- 13:43can use a very cheap API, I don't know,
- 13:45man, I spend like 1 or 2 dollars a month
- 13:48for it to find, like, files in the brain
- 13:50. So it's very cheap, but you start to
- 13:52have that cost, right, of the VPS,
- 13:53indexing, and so on. So when you have
- 13:55the map, the map functions as a summary
- 13:57for your AI. So you have a summary and
- 13:59you can have summaries inside the
- 14:01folders.
- 14:03So you can have a global map and inside
- 14:05each folder you can have a folder map.
- 14:07Inside the skills folder you can have a
- 14:10map of all the skills and what they do
- 14:12and when you last updated them. Anyway,
- 14:14documentation. That is the basic. So the
- 14:17map points to places and helps you have
- 14:19a compass within your structure. Here,
- 14:21JeBrain finds files by subject. So, its
- 14:25search, in a very crude way of putting
- 14:27it, is that it searches for things as if
- 14:28it were Google. So it goes there, you
- 14:30type things in, it indexes, searches,
- 14:32finds everything in a fraction of a
- 14:34second, and it works very well, okay? So
- 14:36, for an example of JeBrain here, what I
- 14:38agreed on with Ana, it will look at the
- 14:39daily note, right, that we mentioned
- 14:41earlier. So, you have the daily notes
- 14:43here. Then it will look at what you
- 14:44agreed on with Ana. It will look at the
- 14:46daily note, contact by Friday, exit
- 14:48clause. So it will look from the most
- 14:50relevant to the least relevant and it
- 14:52will check the file precisely. So it
- 14:54will look there at people/Ana and know
- 14:56that everything is right, that Ana
- 14:57really said that and so on. So it finds
- 15:00Ana in various places, in pending items
- 15:02, in people, in the daily notes, in all
- 15:05places, okay? So once you have the map,
- 15:07you can also have maps, just like the
- 15:09example I was giving, inside the folders
- 15:11. So you can have a map that explains
- 15:13the client folders, what each folder is
- 15:15.
- 15:16You can have a map within the content
- 15:17area about the types of scripts you make
- 15:19, thumbnails, newsletters, and so on.
- 15:21So, the map is a compass. It is very
- 15:23important for any AI to find its way
- 15:24around in there. So, this second step,
- 15:26it is a very cool, very simple step,
- 15:28right? Looking here at the cloud, in my
- 15:30case, I already have JeBrain installed,
- 15:32right? So it will put it here, it's
- 15:33putting it in correctly, blah, blah,
- 15:35blah, and it's installing. Very simple,
- 15:37once again. In 2 minutes here it has
- 15:38already started to install. We'll come
- 15:40back later to see it working. And now
- 15:42let's move to the last step, which is
- 15:44the level where I consider you've
- 15:46reached a very comfortable place. You'll
- 15:48ask me: "But Bruno, is this step the
- 15:50ultimate peak?" Like, I'm a big nerd,
- 15:52just like you, and I want to know
- 15:53everything about a second brain. Is this
- 15:55the place for that? Is this our gold cup
- 15:58, our holy grail? I would say no. But I
- 16:02would say that here it will hold your
- 16:04structure very well. Because to be quite
- 16:06honest, I feel there is no ultimate peak
- 16:08in the AI era. Like, you can install 25
- 16:10different plugins, tools, and brain MCPs
- 16:13, and you might have a brain that works
- 16:15great, or sometimes you might be
- 16:17over-engineering it, right? Which is
- 16:19what we call over-engineering. So the
- 16:21truth is that this structure is the
- 16:22bread and butter that works very well
- 16:24for me. And for my company today, with
- 16:26about 30-something people, with so many
- 16:28things running, lots of context, lots of
- 16:30meetings, it has held everything
- 16:32together really well. So if you have a
- 16:34smaller company or one with less context
- 16:37, man, it's super easy, okay? Brain,
- 16:39storing subjects, people, decisions, all
- 16:41perfectly organized. And now we have
- 16:43this brother here called Roncho. So,
- 16:46Roncho is this guy right here. He is a
- 16:48memory that learns about the user. And
- 16:51the coolest thing is that it is
- 16:53single-player and multiplayer. It means
- 16:54that if you put it in your company's
- 16:57Teams or Slack, it can identify people
- 16:59by phone number, user ID, or email. It
- 17:02doesn't matter how the user is
- 17:03identified, it manages to save specific
- 17:05traits for those users. So it knows who
- 17:07Bruno is, it knows Fabi, it knows
- 17:09Gabriel, it knows Mateus, and it manages
- 17:11to save traits of that personality.
- 17:13Roncho also works within the same VPS
- 17:15where you put the Brain. So you can
- 17:17install both, which is why the cost
- 17:19doesn't end up growing and stays
- 17:21stagnant at around 40 to 50 reais per
- 17:22month. Except that while in the Brain we
- 17:25have subjects, people, and decisions. In
- 17:27Roncho, we have preferences and tone
- 17:29here, right? So, Bruno prefers short or
- 17:32more elaborate answers. Bruno likes this
- 17:34, Bruno prefers that, Bruno decided this
- 17:37, Bruno did that.
- 17:39So it's a place that understands my
- 17:40profile, my patterns, what I do, and how
- 17:42I do it.
- 17:44Roncho also has an open-source version.
- 17:47So, if you come here to Roncho and
- 17:48install it by creating an account on
- 17:50their site, you can, and they even give
- 17:52you some to start using Roncho.
- 17:55I personally used Roncho, okay? I even
- 17:57have my account connected here, but man
- 18:00, I just realized I have 27 sitting here
- 18:03. Man, talk about throwing money away.
- 18:05It probably set it to automatic. It
- 18:08didn't, man, just look. Anyway, at some
- 18:10point in my life, I decided I wouldn't
- 18:12use their API anymore. Man, their API is
- 18:14very cheap. So, the $ they give you at
- 18:16the start, you can use for like a month
- 18:18. It is very cheap. However, once you
- 18:20start scaling and have many people using
- 18:22Roncho, or you've been using it for a
- 18:24long time, those $ 5 start to run out.
- 18:26And then you end up doing something
- 18:27stupid like me, dumping credit in there
- 18:29and it becomes just another tool I
- 18:30dumped money into without having any
- 18:32control over it, right? I dump money
- 18:33into OpenRouter, into Roncho, and so on
- 18:35. And sometimes you just keep leaving $
- 18:36100 there and I just want to die. But
- 18:38the cool thing about Roncho is that you
- 18:40can also get its open-source version. So
- 18:42you have this version here, see, of
- 18:43Roncho, which is this repository, the
- 18:45link is also here in the description,
- 18:46and you're going to take Roncho and
- 18:48install it as well. So it's very simple
- 18:50to install. I'll put it here: "Claude, I
- 18:52want you to install Roncho on the same
- 18:54VPS as Jbrain." In this case, I'm not
- 18:56installing on the VPS, right, Roncho.
- 18:58I want you to test it and explain to me
- 19:01how it works and how you, or in this
- 19:04case, my agents can also pull context
- 19:07from it to work better with me.
- 19:10So, guys, a very simple prompt.
- 19:13The message is: install it, explain how
- 19:15it works, and give me examples of how we
- 19:17can use you.
- 19:18So, in Roncho's case, it has a different
- 19:21issue compared to Jbrain, okay? So this
- 19:23is an important point for us to talk
- 19:25about briefly here. So Jbrain and Roncho
- 19:27, they work like this. Jbrain is
- 19:28on-demand, so it will look for something
- 19:31when you ask it to look for something.
- 19:34Now, Roncho has several work formats,
- 19:37and it even ingests context at the
- 19:40beginning of each session.
- 19:43So Roncho performs this ingestion, right
- 19:45?
- 19:47Roncho has several work formats and has
- 19:49several ways to inject context when it
- 19:51starts a session.
- 19:53So, in this case, I asked Mora how
- 19:55Roncho works, right? So, she explained
- 19:57here, look, let me zoom in for you.
- 19:59Roncho is running here on this Mora
- 20:01server in the Hermes Workspace, blah
- 20:03blah, identity, right?
- 20:04Session. So, the way it's running now is
- 20:07in this model called per-session.
- 20:09So, that's how Roncho turns each
- 20:12conversation into a Roncho session. So,
- 20:14there were 102 sessions, sorry, 62
- 20:17sessions, 602 conversations, okay? So,
- 20:20how does it work, right? How does
- 20:21context enter each session? Roncho has
- 20:24Roncho profile, it has Roncho reasoning
- 20:27, Roncho search, context, and conclude,
- 20:30right? So, a summarized card, the
- 20:32cheapest query, a synthesized answer,
- 20:34right? As Bruno prefers, search, right?
- 20:37How it searches for conversation
- 20:39snippets, a raw portrait of Bruno and
- 20:41that conversation in that session, and
- 20:43it saves a durable preference of mine,
- 20:45right? So, who decides when to query is
- 20:48the agent's rule.
- 20:51So, you define how your agent should
- 20:53query Roncho, whether it should query
- 20:54all the time or not query all the time.
- 20:57Remembering that every time you query
- 20:59Roncho, you decrease your session
- 21:01context because it will inject the
- 21:02session context there.
- 21:04So, you have several ways to define how
- 21:06you will inject it. And Hermesis, for
- 21:08example, in this case, has a Roncho
- 21:10plugin that has three ways to accept a
- 21:12context injection. The first is the
- 21:15hybrid model, so it automatically
- 21:17injects the context and also releases
- 21:20the tools. It has the context, where it
- 21:23only does automatic injection without
- 21:25tools, and it has tools, which is just
- 21:28tools. When to inject, so here is the
- 21:29frequency. So, in this case, it can
- 21:31place it in every turn or only in the
- 21:33first turn, that is, only when starting
- 21:34a session. So, if you are in an infinite
- 21:36loop just compressing the session, like
- 21:38, for example, chatting on Telegram
- 21:40without typing/new, it will only inject
- 21:42it at the beginning of the first session
- 21:44. Then you have the cadence, right, and
- 21:46the quantity, right, dialectic.
- 21:50which is set to eight, right? So, how
- 21:52many turns does Roncho's reasoning run,
- 21:54recall, blah blah. So, when injecting,
- 21:56uh, it has a more technical structure
- 21:58here, right? So, man, this is the Roncho
- 22:00structure, like, in a very simple way,
- 22:02okay? Uh, Roncho broke for some reason
- 22:05here. It just notified me that the
- 22:06plugin was reinstalled. Oh, and also
- 22:08because Hermes had a cool update, so it
- 22:10ended up changing.
- 22:11And so Roncho, man, it works this way,
- 22:13where it has this data it injects into
- 22:15the sessions and such, so it is more
- 22:17active than JBrink, which was more
- 22:19reactive in its searches.
- 22:21And finally, man, I built this little
- 22:23engine here. It's a very simple little
- 22:26engine that I call the Context Ledger.
- 22:28So what does it do, okay? The context
- 22:31ledger identifies where the signals come
- 22:34from.
- 22:35So it's like it keeps grabbing keywords
- 22:37every day, okay?
- 22:38So it takes here, for example, and says
- 22:41: "Look, on the 26th, 10, 14, in the
- 22:43WhatsApp source, Ana, a person sent a
- 22:45contract today, the reference is on this
- 22:48date here, which is where it pulled the
- 22:50conversation." So it saves it here. So
- 22:52it identifies the signals.
- 22:55So it puts WhatsApp, email, granola,
- 22:56right, which is my transcription.
- 22:59So it's always picking up things that
- 23:00are tags, right? What is the next step?
- 23:03Accountant, invoice, did it arrive, was
- 23:05it invoiced? So they are several
- 23:07keywords in a log.
- 23:11When it hits 90 days, it archives the
- 23:13last 30 days and keeps saving so it
- 23:15always has these 60 days of context. So
- 23:18it's always pushing 30 days, archiving
- 23:19the last month, push, archive the last
- 23:21month, and then it goes on saving in the
- 23:23archive. So if one day I need to search
- 23:25for some information from this Ledger,
- 23:27Jebrain will find it. And then inside
- 23:28the agents, right?
- 23:30So inside this file, which can also be
- 23:32your cloud.md, for example, it explains
- 23:35that every time it loads my brain, it
- 23:37loads my structure, it has to load this
- 23:39ledger so it knows where to find things
- 23:41.
- 23:43So, it's just a simpler way for it to
- 23:45locate things faster and have context of
- 23:47where everything came from.
- 23:49Because sometimes it has to keep
- 23:50searching, if your brain is too big and
- 23:53you index, like, files of 30 employees,
- 23:55of 50, 100 employees, it starts to get
- 23:57very heavy.
- 23:58So this ledger helps to reduce it a bit
- 24:00, but it's just a thesis, I'm not even
- 24:01sure if it makes sense, in my view it's
- 24:03a thesis.
- 24:04Anyway, if you have a suggestion, or
- 24:06disagree, comment below. I love learning
- 24:08from you, okay? So, guys, this is it.
- 24:10This is the third step. And I'll leave
- 24:12you with a challenge, right? What is
- 24:13your step today? Are you on a free
- 24:16search level or one that knows you? For
- 24:19you to get there: GitHub, Gbrain, Maps,
- 24:22Gbrain plus Roncho, and the Context
- 24:25Ledger. It's just an extra bonus. And
- 24:27then you decide if you want to create it
- 24:28or not. You can create it again from the
- 24:30transcript of this video and it will
- 24:31work, okay? So folks, this is my video
- 24:34explaining how you can move from the
- 24:36first step to the third step, how you go
- 24:39from a free system that works very well
- 24:42and functions as a backup to a search
- 24:44system, to a system that knows you,
- 24:46being fast, being mobile, without losing
- 24:49context, without ever forgetting. And
- 24:52this same dynamic works single-player
- 24:54and multiplayer. So, if you install this
- 24:56in your company, your whole team can use
- 24:58this structure too, okay? Guys, more
- 25:00than 20 people asked me to record this
- 25:02video. So, if you made it this far, if
- 25:03you liked it, man, leave a like.
- 25:06This month of October is one of the
- 25:07months where I'm getting a lot of views
- 25:09on my videos and few subscribers, man.
- 25:12A lot, man. I enjoy talking to you so
- 25:15much. Follow the channel, man. And
- 25:17that's it. Any topic suggestions you
- 25:19have, comment here on the video. It will
- 25:21be a huge pleasure to be able to record
- 25:22another video. I'll see you next week.
- 25:24We're in this together.
About this transcript
This page contains the full transcript of Como construir um segundo cérebro (guia definitivo) by Bruno Okamoto, generated from the public captions YouTube serves with the video. The transcript has 5,159 words across 750 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 choose a caption track available on the video.
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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.