Most AI Second Brains Are Built Wrong

Author: Leon van Zyl · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Most AI Second Brains Are Built Wrong · Channel: Leon van Zyl · 10:15 · Watch on YouTube

Topics: AI agents, Claude, RAG, search and information retrieval

Summary

The video demonstrates how to build an AI second brain using Progress Agent RAG, then connect it to an assistant such as Claude or ChatGPT. The setup uses a knowledge box as a searchable database instead of relying on markdown files, which the presenter describes as inefficient and difficult to scale. An agent can upload videos and preserve their titles and source links, while an MCP connector lets Claude search the stored information and return relevant answers with sources. The tutorial walks through creating the knowledge box, setting up an uploader, adding NASA videos, and connecting the database to Claude.

Key points

The presenter argues that markdown-based second brains are slow and do not scale well for agents.
Progress Agent RAG stores information in a searchable database that an AI assistant can query.
A knowledge box is the collection where uploaded files, videos, and other sources are stored.
The tutorial creates a reusable uploader script with an agent that can execute code.
The uploader is instructed to retain titles and source links and avoid duplicate entries.
The example adds 15 NASA Science Cast videos, including “Dark Lightning,” to the second brain.
An MCP server endpoint connects the knowledge box to Claude through a connector.
Claude can search the database and compare information while linking to the original sources.
The presenter says the connection works in both the Claude desktop app and the Claude website.

Chapters

Start	Chapter	Summary
00:00	Demo and why use RAG	The presenter demonstrates Claude finding a NASA video in a second brain and explains why a searchable RAG database is preferable to markdown files.
02:07	Create a Progress knowledge box	The tutorial shows how to create an account and configure a knowledge box, including its name, region, embedding model, and optional anonymization.
03:46	Set up an uploader and add videos	An agent prepares a reusable upload script, then finds and uploads NASA videos while keeping their titles and source links.
07:35	Connect Claude and query the second brain	The presenter connects the Progress MCP server to Claude, tests a search across two videos, and notes that the connector is available on desktop and web.

Quotes

The majority of second brains seem to rely on markdown files, but this is simply not scalable, and it’s way too slow. — 01:05

CoWork will add the files, Progress will turn its files into a searchable database, and Claude chat will search the database whenever we ask questions. — 02:07

We only have to do the setup step once. — 05:56

This gives Claude access to three different tools. Batch get documents, get document, and search documents. — 09:10

Concepts and entities

Second Brain, RAG, Progress Agent RAG, Claude, Claude Desktop, ChatGPT, MCP, NASA Science Cast, archive.org, Leon van Zyl

Full transcript (original language)

00:00 In this video, I’m giving Claude a second brain. You know when you remember something useful from a video, but you can’t remember which video it was. I just want to ask Claude about it. It should go and find the information I need and give me that original source. In this video, I’ll show you one of the easiest ways to set up a second brain. This works with any AI assistant as well. You can use Claude, ChatGPT, or any agent, whatever you want. We will be using Progress Agent RAG, which is really easy to set up and beginner-friendly. Let me show you a quick demo. So, currently I’m using

00:32 Claude Desktop, and if I have a look at my connectors, you will notice I’ve got my second brain connected over here. I’ve uploaded a ton of NASA videos to my second brain. And if I want to access any information from my second brain, I can simply ask Claude, “Hey, search my second brain. I remember a NASA video about thunderstorms creating antimatter. Which video was it? Explain the connection and link to the source. Use only my second brain.” And cool, it used the second brain, and it’s saying the video is the NASA Science Cast episode

01:05 Dark Lightning, which it was. And I’m not going to read all of this, but this was exactly the information I was looking for, and this also includes all of the sources. This video is going to be a lot of fun, so let’s dive in. Let me just explain how this solution works and how it’s different to other second brain solutions out there. The majority of second brains seem to rely on markdown files, but this is simply not scalable, and it’s way too slow. It’s simply not efficient for agents to crawl through markdown files to retrieve and

01:36 store information for you. Instead, our second brain is going to use RAG, or retrieval-augmented-generation. Now, RAG is nothing new. It’s been around since agents became really popular, and it’s a blazingly fast solution for uploading information and querying data as well. Now, don’t worry too much about how all of this works behind the scenes as Progress will handle the database for us. We just know that these special kinds of vector databases are blazing fast to query. And if we ever want to retrieve that data,

02:07 we can simply ask Claude to read from our second brain, and it will go and retrieve the results from our Progress database. So, basically, CoWork will add the files, Progress will turn its files into a searchable database, and Claude chat will search the database whenever we ask questions. In order to get started, we will need to create a Progress account. So, I’ll leave a link to the website in the description of this video. Or just go to rag.progress.cloud. Then we can click on create new account and just fill in this form. Right, after signing in, the first thing we need to

02:39 do is create a knowledge box. Now, this is basically a collection of data. So, all of the files and the videos, anything we share with the agent, will be uploaded to the knowledge box. Now, because I’ve already got one, it kind of just defaulted to my existing second brain. So, all I’m going to do is just go to my profile. Let’s go to knowledge boxes. And then from here, I’m going to create a new knowledge box. I’ll just call this one YouTube second brain. And for the description, I’ll just enter my second brain. We can also select the

03:12 region. So, I’ll just do USA because this is using rag, we can also select the model that we would like to use for the embeddings. So, we can select this by language. I’ll just go with common languages. But if you want, you can also go to external trusted partners and select the embedding model like OpenAI’s models, Google, Gemini, and Hugging Face models as well. I’ll just go with their default, which was common languages. And what’s really cool is we can apply GDPR compliant anonymization as well, but this is optional. I’ll just leave it

03:46 off. Let’s create our knowledge box. So, let’s click on our newly created second brain. Then what we can do is go to resources list, and this is where we can see all the data and sources that we’ve loaded into our second brain. Of course, at the moment, this is completely empty. So, let’s fix that. Let’s see how we can upload data to our second brain. For uploading data to the knowledge box, we will need to run an agent that’s able to execute scripts. So, if you are using Claude, you can simply use code work, or if you are using ChatGPT, you can switch to work mode as well. Agents like Hermes

04:20 agent will not have this limitation. It is already able to execute code. Then, let’s also assign a workspace or folder. I’ll just create a new folder called second brain on my machine. I’ll allow access to that folder, and this will simply allow code work or ChatGPT work, or whatever you’re using, to write the scripts necessary to upload data to the second brain within this folder. Then, let’s set up this environment. I’m going to pass this prompt to code work. You can find all of the prompts I used in the description of this video. Set up a

04:52 reusable uploader for my Progress knowledge box called second brain. And if you want the exact name of your knowledge box, simply click on home and copy this name over here. It should upload videos, save and update notes, keep sources links, and avoid duplicates, and use this guide. This is simply linking to the official documentation from Progress. And my knowledge box API URL is Then, we can get that URL by clicking on these three dots. Let’s go to developer integrate. We can copy this nuclear DB API

05:22 endpoint, so this first option over here, and paste that into this chat. Then, I’ll also add my API key is And look, what I would recommend you do is just kind of get the agent to set up all of this without you passing the API key, and then you can set the API key afterwards. But, for the tutorial, I’m just going to pass it in the chat and delete the key after this recording. Now, in order to get our API key, what we have to do is go to advanced API keys. I’ll just call this Claude. For the role, let’s set manager, and let’s

05:56 add this key. We can also set an expiry date. Let’s click on add. Let’s copy the key and pass that into the chat as well. And that’s really it. And Claude will now set up everything we need to upload data to our second brain. We only have to do the setup step once. After this, we’ll simply ask Claude to remember something or add something to the second brain, and it will reuse these existing scripts. Right, so Claude actually just created the script, and what I’m going to do is ask it to pull videos from archive.org. So, if you want to follow

06:29 along, you can just search for NASA Science Cast. This will bring up a ton of very cool NASA videos. Back in Claude, what I’m going to do is let’s ask it to find 15 downloadable NASA Science Cast videos on archive.org, including dark lightning and hidden magnetic portals around Earth. Upload only the dark lightning MP4 to the second brain. Keep its title and source links, and confirm that it’s searchable. Let’s send this. While Claude is working on this, I do want to mention that you can upload documents and references

07:01 yourself. You can simply click on upload data, and you can upload files, upload folders, links, text documents, site maps, and Q&As. But of course, we want our agent to maintain our second brain for us, right? And look at that. We’ve got our very first resource. So, we can actually click into this resource to see what it’s all about. We’ve got a ton of metadata here. If we click on text fields, we’ve got the URL and the name of the file, and the file field contains the entire transcript from the video, and all of this can be queried by Claude. So, since this is working, I’m just going to tell Claude

07:35 upload the remaining 14 videos to my second brain. Keep the title, source links, skip duplicates, and tell me when all 15 have finished processing. We’ll skip ahead to when this is done. Right, now all 15 of these videos have been uploaded to our second brain. And if we ever want to upload anything else to the second brain, it’s just a matter of selecting that folder, clicking co-work, and asking your agent to add whatever. Now that we’ve set up our second brain and we managed to get Claude to populated, we need some way to get Claude to retrieve information from the second brain as well. And this is really

08:07 easy to set up. So, let’s go back to our second brain in progress. Then let’s click on these three dots, go to developer/integrate, and now let’s simply copy this MCP server endpoint. And I’m sure you know exactly where this is going. We can simply go to chat or co-work. At this point, it really doesn’t matter. Let’s click on this plus button, let’s go to connectors, and manage connectors. Then let’s click on add. Let’s give it a name like second brain, and space in that MCP server URL. Let’s

08:40 click on continue. For authentication, we can just leave it on sign in now. Under OAuth client, let’s select register automatically. And that’s it. Let’s click on add. This is the same process that you would follow if you are using ChatGPT as well. And if you are using something like Hermes agent, those agents will automatically set up the MCP server for you. All we have to do now is authorize access to progress by clicking on connect. That will bring up this URL, so I’ll just click on continue connecting.

09:10 Then let’s click on allow, and done. Our second brain is now connected. This gives Claude access to three different tools. Batch get documents, get document, and search documents. This is everything we need to query our database. So, we can actually test it with a prompt like, “Search my second brain, compare how magnetic fields move particles in dark lightning and hidden magnetic portals around Earth. Keep it simple and link both sources. Use only my second brain. I’m just going to ensure that my second

09:42 brain is enabled, which it is. We do not have to select co-work. We can leave it on chat. It will still work. And let’s send this. And there we go. We get our response back. Both are NASA science class in your second brain, which is correct. And they describe the magnetic field doing two different jobs. And for both of these, we do get the sources as well. I should mention that I am using the Claude desktop app at the moment. But this will also work in the Claude website, by the way. If I go to claude.ai, so the actual website, I can go to connectors. And check that out. The second brain is available in the web

10:15 app as well. So the moment you set up in either the web app or the desktop app, it automatically syncs and you get access to your second brain from anywhere. And now you have your very own second brain, and you can just use co-work to upload more information to the second brain, and use chat to query any data. If you like this video, then hit the like button and subscribe to my channel for more authentic and entertaining tutorials. Thanks to Progress for sponsoring this video. I’ll link to their website in the description of the video. I’ll see you in the next one. Bye-bye.