# oracle/transcripts/chunks/c06.mp3
# model: tiny | lang=en

[0.0s] custom-wise and personalized here. That is unbelievable. So if we go inside here and we're like,
[4.3s] okay, create a image of a cat eating biscuits and we'll delete this bit here, we'll change that.
[11.8s] We'll get rid of that. You see how it's got this aspect ratio as well? So I didn't even ask you to do that,
[17.0s] but code X4 ahead because that asks it to figure out. That's why I quite often prefer just
[21.4s] giving the AI to figure out because it has better ideas and it has more time to give attention to this
[26.7s] than I do. Like if I've been coding this out for hours, I'm going to get tired, I'm going to get
[32.2s] distracted, is going to be difficult to focus at times. You might get frustrated sometimes.
[37.1s] Well, where's if you give the decisions to a frontier model like GPT-5 on 6, it can do it all for
[43.2s] you and handle it as you go along. So now we can say, okay, I want to just go with a cinematic image of a
[49.6s] cat eating biscuits. We've got the prompt, we've got the aspect ratio with cinematic, we'll hit generate
[54.7s] image and that may not work, but so far it's actually looking really good, really impressive actually.
[61.5s] And so one thing that might happen here is if we click from image to video to homeization, I wonder if
[68.0s] it's going to stop? Yeah, see how we click on that and then it's stopped generating. So we need it to be
[73.0s] able to run in the background. That's another great feature. And you'll see the more you test it, the more you'll
[77.4s] see ideas from improve it and more things that will come up that are kind of like bugs by this. And so
[83.1s] what we want to do here is just say, okay, inside the custom workflow image generator and this is exactly
[89.2s] why you want to rename the conversation. So if you right click and rename the conversation, you can see exactly
[93.9s] what you're working on. It keeps makes it easy to keep track. We can go inside here and say, based on the image
[100.2s] you, if I generate an image, can you just make sure it still runs in the background whilst I'm moving around?
[107.1s] And also what we've seen here is that it's actually tested out and it's found some bugs that we didn't
[112.6s] even know about. So it's turned a 500 error and it's uncovered a real bug there which is fantastic
[119.4s] as well because it's still in the bug checking for me. But overall it's looking very, very promising
[125.0s] that's beginning to work. I mean, a bit of we go back to the app here and we're like, okay,
[131.8s] create an image of a cat and then we just hit generate image. That should actually work without us
[137.8s] switching between them. But let's see. So the mission control, we have the different agents built in.
[143.0s] We have the studio of custom workflows as well. But this is beginning to look and form something
[149.8s] that is actually amazing. And you can see here that we've got the agentic memory of all as well.
[157.0s] So this is the system that we've used and as you want to add that as a tab. So I'm going to say,
[163.0s] add the memory system as a tab. So the I can easily see the memory has been added and see what's
[170.0s] been done and also visualized the inside like a beautiful knowledge graph and we can hit into here.
[174.6s] Now it's beginning to run out of work usage as you can see because I've been running on fast mode and we've
[179.8s] created so much stuff started by side. I'm going to say figure this out. But the 8020 of building the
[185.5s] agentic operating system is being done as you can see. And so all of this is working in the background.
[190.5s] You do need to be careful with the token usage as well because you can easily run out of tokens.
[194.6s] If you're wondering how to check your token usage, you can go inside the advanced section here and you can
[200.6s] switch the effort levels and speed and everything else. As you can see how many tokens you've used side by side.
[207.4s] And you can also click on reset usage as well as an option. Now this is beginning to code out each part as well. Nice.
[217.0s] Memory system should be added in and we hit the limit. So there we go. We hit the usage limit already.
[226.2s] We can actually click on reset usage. I'll be quite interested. I know if we can and I'll. Oh yeah we can.
[231.2s] Of course you can do a weekly reset usage limit as well. I didn't know that.
[237.0s] That's pretty good. And then we'll go out weekly usage limit that resets on the 18th which is quite a few days ago.
[242.6s] But yeah and also when you're being out with this. If you want to reduce the amount of tokens
[248.6s] because this can be come up problem especially if you're creating some of the things in parallel using the
[254.4s] GenteColoring system, then you've got a few options. So you can switch your fast mode. You can switch off the
[259.9s] intelligence so you can actually dial down the intelligence level. You can also change and click on advanced
[266.8s] and then you can switch between the models. So if it's just something really basic then you could switch
[272.0s] the model to like 25.5 and that would use less tokens. You can change the effort level to light or medium
[279.0s] and you can change the speed standard. Now when you're building all this out,
[283.4s] the more you reduce the amount of tokens you use by dialing down the effort dialing down the speed,
[288.7s] dialing down the model, the worse the outputs will be. As always that balance with token optimization
[294.6s] creating something that you really like also using the plan.