ChatGPT Dots: como usar de forma eficiente (tutorial completo)

Author: Matheus Battisti - Hora de Codar · 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: ChatGPT Dots: como usar de forma eficiente (tutorial completo) · Channel: Matheus Battisti - Hora de Codar · 16:08 · Watch on YouTube

Topics: AI agents, Task and workflow automation, AI models

Summary

The video reviews OpenAI’s Dots as an AI personal assistant that can work asynchronously, use connected apps, handle files, and communicate by voice. The presenter demonstrates tasks such as checking email, creating content from subtitles, producing a PDF, and responding to a Slack request. He also highlights current limitations, including restricted access to Codex conversations, uncertainty about saving files to a computer, and the inability to end a call by voice. Overall, he sees potential in Dots but considers it an early, incomplete product that needs stronger integrations and more reliable actions to become useful in his daily work.

Key points

Dots can continue working after the user closes the chat and may notify them when a task is complete.
It can search the internet or connected services, use files, and carry out some actions, while requesting approval for certain tasks.
The presenter connects Dots to email and demonstrates asking it to identify urgent messages.
He uploads video subtitles and asks Dots to suggest related content and prepare a YouTube description with chapters and a PDF.
A voice call lets the presenter ask for progress updates, but Dots cannot reliably summarize all open Codex conversations.
Dots can receive a Slack request and return a summary in the Codex chat.
The PDF is attached to the chat, but saving it directly to the Downloads folder is not confirmed.
The presenter values Dots’ potential for managing projects asynchronously, but says it needs better integrations and more dependable capabilities.

Chapters

Start	Chapter	Summary
00:00	Introducing Dots	The presenter introduces Dots, describes its intended computer and app access, and cautions that it still feels like an early release.
03:11	Security and the move to a live demo	He explains some blocked or approval-required actions, then briefly promotes a separate training course before turning to Dots in Codex.
04:45	Connecting apps and creating content	The presenter demonstrates connecting email, checking for urgent messages, and asking Dots to create content and a PDF from video subtitles.
07:23	Voice calls and Codex access limits	During a call, Dots reports on its work but cannot reliably access the full list and history of Codex conversations.
10:59	Task status and call controls	The presenter reviews Dots’ activity indicators and file-delivery status, then notes that it cannot end a call by voice.
12:00	Potential and current shortcomings	He discusses the possibility of managing and checking projects remotely, while emphasizing that Dots still lacks useful integrations and reliable functionality.
15:07	Final assessment	The presenter concludes that Dots needs to become effective in real-world use, since its current limitations may keep it out of his daily routine.

Quotes

It’s a ChatGPT that keeps working after you close the tab. — 01:03

Right now I’m working on that PDF of your last video, five clips, chapters, and posts. — 07:23

I couldn’t read or send a message in that conversation already open in Codex. — 08:58

For now, I can’t end the call for you. — 11:29

So it doesn’t fly, it has to be effective. — 15:37

Concepts and entities

Dots, OpenAI, Codex, ChatGPT, Slack, Astra, Grokbot, Jev

Full transcript (original language)

00:00 OpenAI released Dots, their version of the Grokbot. So, it’s basically software that will have access to our computer, the internet, and will be able to perform actions. As we chat with it synchronously, it will have access to Codex projects. It’s quite interesting. I’ve been testing it a lot and I want to give you a quick summary and show what I’m already doing with it. So, check it out. Well, guys, I found the project very interesting, just to give you that disclaimer, but I think it’s still

00:32 very much in alpha, right? It doesn’t work as well, let’s say, as the Grok Bot. It has its limitations, throughout all my tests. But I believe it’s a good starting point. It already does some of the tasks I need from a personal assistant, something that’s more interconnected with what I use. I don’t use the Grok models there, I use Claude and Codex more. So for me, it’s interesting to have this guy, to have this alternative, okay? So I put

01:03 it here that it’s a ChatGPT that keeps working after you close the tab. Basically, it will be acting there until the task is finished. It will notify us in several possible ways, okay? Right here, right? If we were to compare it with ChatGPT itself, the chat goes there, searches for what it needs to search, and returns the answer to us. Dots, on the other hand, will have the ability to understand what we need, to search the internet if needed, search our computer too if necessary,

01:36 and execute commands, either on our computer or in apps we connect to it. For example, the Google Suite, Slack, or something else, and then test if necessary, ask for approval, or deliver it, depending on what we requested from it. So it’s like it’s an employee of ours, right? We can think of it that way. It has its own computer in the cloud, so it can execute various things without depending on our computer. It

02:07 connects with more than 4,000 apps, right? ChatGPT also already does this. We can plug in our email, plug in a spreadsheet, plug in Slack, and others it accepts, and it works with them in an interconnected way, which is quite interesting in this personal assistant model. You can call it, and I’ll even show you how in practice. And it learns both in the chat and on the call. It saves this in its memory. So it has a “ make Hermes” vibe and it’s still running on Astra, right? Well, guys, it

02:39 has some security locks here, which is really cool, right? It can do a lot of things on its own, we can give it autonomy, okay? But for some things, it will ask for approval, for example, to send an email, to post on Slack, this will be requested from us so we confirm it can proceed. And other things are totally blocked because it will be on our computer, right? So there could be a risk factor if it had access to everything, right? So, for example, if

03:11 we block something, right? If we proactively ask it to block, it won’t do it. And also some actions, like changing a password, it cannot perform either, okay? So that’s quite interesting, because it gives us a little more security. But enough talking, right, guys? Let’s go see it in practice. First of all, guys, we have the waiting list for the Jeev course here. It launches this coming Monday, okay? October 5th. I will send an email to those who signed up on this list on this page, which will be in the

03:43 pinned comment, and those people will have a special launch condition, okay? So, if you want to participate, if you want to learn Jeev, I recommend you pre-register, at least to get the founding class discount. Jeev, guys, is an agent. We have several videos here on the channel. I recommend that you at least check it out, because it’s something different from what we are used to seeing in terms of news in the AI world. I think it’s totally worth it . Even OpenAI has already launched agent APIs, right, which are coming soon, and everyone is already saying it will kill Jeev. But look at how this

04:13 has already shaken up the AI world, right? OpenAI itself talked about this product at Dev Day. So that means it is indeed promising, and the ones who, let’s say, brought this into the commercial world were the Jeev team. And I believe that the pioneer always has the most evolved product. So it’s worth learning and then perhaps migrating to the competitor’s version and so on, but the operation will be the same. So here’s the invitation to participate, sign up on the page here and master Jeev, alright? Well,

04:45 speaking of agents, what happens? You can run it on the web version, right, on ChatGPT, but for me, the most powerful way is here directly in Codex, okay? In Codex, it will stay right here in the corner when you go to the home page, right? It will appear enabled here for you to create your bot. Mine is already created; I configured it and have used it quite a bit. And I’ve seen several ways to use it. The cool part is that it has connectors there, okay? So we end up connecting it to various apps. For example, these are my

05:15 connectors here, right? So I can already talk to it and access my email to get an idea of what happened today, or ask it to summarize. I’m going to ask it here, were there any urgent emails today? And since it’s already connected to my email, it will read them and report back. Oh, it’s asking here if I’m recording. I mentioned to it that I was going to record the video and such, so it saved that to its memory. Just to show you all. I am, but go ahead, just don’t show any confidential details. I’ll say that to

05:49 it, right? It actually made me a bit nervous to check now. I didn’t find any urgent emails. Alright. But it’s cool , right? That it saved what we talked about in the call. This is very powerful, guys. Now, let’s bring in something from my daily routine, right, of content production. I’ll paste the subtitle of one of my videos here that did well, right? I’ll ask it here for ways to create other content related to it, okay? Also a description for

06:22 YouTube with chapters, which is quite interesting because it would speed up my work instead of me having to add timestamps. So I uploaded the SRT file, and I can ask it, and that’s it, it will keep working here autonomously, doing all its research and file creation if needed. I’ll add here, I want a PDF, and then we are usually notified by Codex, okay? Notice that

06:52 the conversation is asynchronous, right ? It doesn’t get stuck in a queue, right? The message I send, we just keep talking as if it were WhatsApp. Now I’m going to call it, guys, because I thought this call feature was really cool, just so you can see how dynamic it is, right? It’s in European Portuguese, but the response is very quick. Look, it seems to be calling here. Actually, that even failed in OpenAI’s video, during the live demo

07:23 they tried to do, but it worked here when I tested it. >> Hey, Mateus, good to see you again. >> Oh, it’s talking right there, look. I’m recording a video here, I want to know what you’re doing for me right now. >> Sure, let me update you. Right now I’m working on that PDF of your last video, five clips, chapters, and posts. I haven’t delivered the final file yet, but when it’s ready I’ll send it here in the chat. >> It’s still in production, okay? As soon as it’s validated, I’ll send you the file here. >> Can you send a summary of the projects we are working on here in Codex?

07:56 >> Okay, wait a second while I pull that up for you. I can certainly put together a summary, but I don’t have a reliable view of all the open conversations in your Codex. I will check what is accessible and separate what I can confirm from what I would need you to show me. I’ve sent in the chat what I can confirm right now. I still don’t have reliable access to the list and history of all Codex conversations, so I can’t summarize all the projects without you providing the conversations or links. >> Look, guys, that’s what I’m talking about, it doesn’t have direct access to Codex, right? That, to me, is something

08:27 that has to change, because then it will be more useful. You heard it saying there that it can interact in a certain way, but it can’t send a message in the project, it can’t see the conversations. I believe that this will change in the future. It will have all this interconnection and this intercommunication between our DOT agent and the projects. And only then will it make our lives easier. Can you confirm this, HDC?

08:58 >> Let me check carefully to give you an accurate answer. I confirm the limitation we saw here. I can start and track some connected tasks, but I couldn’t read or send a message in that conversation already open in Codex. About this changing in the future, I also think it would be useful, but I don’t have confirmed information to promise when or if that integration will arrive. >> And another cool thing, guys, is that we can talk to it on Slack, right? And then it answers here in our conversation. For example, I have a server here, right? And then I’m going to send it a message here. Put here,

09:28 summarize the OpenAI Dev Day for me in Codex. Then it receives the message. >> Oh, and your video PDF is ready and is already attached to our chat. Five clips, description with chapters, and the posts. I haven’t published anything . >> Look, we can even double-check that with him, right? Did you receive a message on Slack? >> Hold on, let me check. >> He did receive it. Yes, you asked for a summary of the OpenAI DevDay to be delivered here in Codex. The search is starting, the summary isn’t ready yet.

09:59 >> Look, now I’ll check the PDF he created here. I saw that he also has the ability to create files, right? HC, where did you save this PDF? It’s on my computer. >> Let me check that for you now. >> Look, here is the PDF he created. >> It’s attached to this chat conversation . It wasn’t saved automatically to a folder on your computer. Just click the attachment to open or download it, and then you choose where to save it. >> Can you put it in downloads for me now, please? Let me check on that. I’ll try to save a copy to your downloads folder and confirm the file name. This request has been sent, but there’s no

10:29 confirmation yet that the file appeared in the folder. I’m still trying to place it there. I’ll let you know as soon as I verify it. >> Look, over here guys, he’s performing the action, see? Do you see it here? He’s trying to do something. Then the load appears here, right? And he continues doing the other one. So it’s asynchronous. There are several tasks running at the same time, right? Look, this was the DevDay summary, right? So, a few releases, right? GPT-4 Turbo, the more complete Codex, the Dots, and so on, right? I’ll just let him finish here.

10:59 >> The DevDay summary is already in the chat with official links. >> I already sent you the DevDay summary in the chat regarding the PDF. I still don’t have confirmation that the copy was saved to downloads. >> And here, guys, below his avatar, he shows everything that’s happening, right? where he’s connected, which files he changed or made, what he’s working on, his activity there. So, we can follow this very complete summary of the statuses, right, that are

11:29 happening in Dots. Another thing I thought was bad is that he also doesn’t have an option to hang up here, for example, “Hey HC, end our call.” Can you end the call? >> Let me check that really quick. For now , I can’t end the call for you. You need to end it using the call button, and yes, it would be much more practical to be able to ask for that by voice. >> It can, right? I think it’s still limited in some actions, right? But it will probably evolve. Hey guys, I turned it off because it was talking over my voice, you know? But honestly, guys, what do I see in it? I view it

12:00 positively because it has this interconnection between various software I use in my day-to-day. It is in the Codex, which is also part of my projects. So, as it advances, with the development of Dots unfolding and it being able to interact with projects, perhaps we can do remote development across several systems, deploy, see, test in real time, ask it to do this totally asynchronously, without needing

12:31 to be at the computer and getting its real feedback via this call or even through the chat. That would be very interesting, at least in my case, since I already see the OpenAI models neck and neck, let’s say, with Cloud Code, right? And what I’m really missing is having this power to not be in front of the PC and still be able to move projects forward safely, right? Not just asking it to do something and that’s it, but being sure it will execute, be able to verify, and give me

13:03 the green light. That would be very interesting and I think they are heading towards that end. Dots already uses Astra, so that’s a big step forward. To me, Astra is one of the best models out there, right up there with Opus and Claude, right? And that also brings a lot of potential to what it can accomplish, right? For it to perform with more assertiveness. This lack of functionality and integration also bothers me a bit. But as I said at the beginning, it feels like it was an alpha release, a bit rushed. But it has that light at the end of the tunnel

13:35 that makes me think this will be very useful when a version comes out that integrates more things and allows me to perform more actions through the call. That will be really good. And here in “ more” we have other features, right? We can connect with more plugins too, or add files and such, like a chat conversation, but with all the potential that Dots brings. I personally don’t use Slack, so it doesn’t make much difference to me, but

14:07 I know it’s widely used in business environments, but the call feature works great for me. And if it evolves, I intend to use it via calls and be able to manage my projects and get that real-time feedback. I really like how the call flows, it seems very dynamic and it actually responds and so on. I did a few with it there and it works really well. But guys, that’s it, there’s not much else to show. It’s a personal assistant. with limited powers that runs on a good model and is backed by a company at the top of the AI field

14:37 . So, we just have to wait and see what happens in the next chapters. And to wrap up, I invite you all once again to participate in our Jeff training, okay? The link to sign up will be here in the pinned comment. On that page, I explain more about the training, guys. And it’s this coming Monday, okay? So sign up now to secure this promotional price. Now I want to know from you, what did you think of the dots, right? If you have the same opinion as me, you didn’t like this launch at all. For me, that’s

15:07 it, right? I don’t even want to defend OpenAI here. I’m really liking their AI model and I believe they are cloning what’s working in the AI world. For example, Grokbot, which we’re speaking well of, Jeff with decision API, and that other feature from Cursor that they also released something similar for. So that’s interesting. We have more competitors, we have a good company behind resources we use in our day-to-day. Now it needs to be truly useful for us, because half-baked like

15:37 this won’t be of any use, nobody will end up using it, the product will just die and it won’t have any real utility, right? At the point it’s at, it doesn’t seem like it will fit into my daily routine, right? It seems easier to just go into the codex and do things than having to open another tab and try to do it only to see it doesn’t work, right? So it doesn’t fly, it has to be effective. So we have to see if it will evolve and then actually have real-world use. Well, I’ll leave it there, guys. Don’t forget to subscribe

16:08 and leave that like. Best regards and see you in the next video.