athena-oracle/transcripts/chunks/c10.txt

# oracle/transcripts/chunks/c10.mp3
# model: tiny | lang=en

[0.0s] He was like, you might be tempted to have Claude and Codex work on the same thing at the same time.
[7.0s] I personally wouldn't recommend that.
[9.5s] So if you are building out with an agent and you're trying to work on the agent,
[13.8s] as you're trying to improve it day by day, trying to make it better as a system,
[18.0s] the thing that I'd recommend here is that you just use one agent.
[22.0s] So for me, for example, I'm using Claude right now and I use Faber 5 to help me,
[26.0s] you might prefer using chatchipy to you might prefer using Hermits, but seek to one.
[30.0s] Because if you have multiple agents, it multiple different apps, number one that gets really messy when you're building out an agent offer in system.
[38.0s] And number two, they could like overwrite or delete something that the other agent is building.
[45.0s] So for example, Claude is building out something that's just broken.
[48.0s] Chatchipy might be like, let's remove that from the code.
[51.0s] And then all of a sudden you've got two agents kind of interrupts any chance work and that can get super messy.
[56.0s] Wouldn't recommend that.
[58.0s] One thing that I haven't tested, that would be very interesting in testing mode, is with Codex is very easy to add plugins.
[66.0s] Now, what would be very interesting here is if you could add those plugins inside the agent or S as well.
[73.0s] So that's an example that we've got email and Gmail plugged into Codex, right, into Chatchipyty work.
[80.0s] So you can send the email address, you can manage your email.
[83.0s] So you can draft replies for us and if it else.
[85.0s] Now, what would be very interesting is say to Codex, a build out this scheduled task into my agent or S, so that we have the email agent inside there.
[99.0s] I'm not going to do that right now, because I think it's a task that we'll probably take about 20 or 30 minutes to really perfect.
[104.0s] But that would be interesting to take for the plugin section from Codex, and start adding that into your agent or S, so that you can automate and build anything as you want right there.
[115.0s] Other things that I would recommend.
[118.0s] Number two is you might be tempted to put this on a website, for example, like a lot of our websites we use with NetLefite.
[127.0s] I wouldn't do that just because again, like I think you need to keep it secure in everything else.
[132.0s] Also, if you want to reduce the amount of tokens, you can also use stuff like Free APIs and you can use local models.
[140.0s] So if we go to the local section here, we can use local models from Olammer or LM Studio, and this is particularly good for like the tasks that a quite time consuming, but don't need frontier levels of effort.
[155.0s] So let's say for example, we're looking at all of the models on Olammer.
[161.0s] I'm like, okay, or NetLef, or NetLef looks awesome, right?
[165.0s] Actually, it's a really good tool. I've test it out.
[167.0s] Persona, we have the local lead board on Goldiebench.
[171.0s] We've all the local models, and you can see what we've created in T.R. web, Formula Finals.
[175.0s] And so you might be tempted to build that into your agent or S as well.
[179.0s] And I think it's actually pretty interesting to build out a local code.
[183.0s] And so what we have over here is we have this local section, and we have the preview of F and created what the works face of our previous generations, and we can use local models in SOT System 2.
[195.0s] And that's another way to use it for free, because then you are a combination of like Omniru, which is free way to use like 90 models, and it kind of orchestrates across them.
[204.0s] You have HY3, which is another free API, you can plug into the system, and then you have local models, which are basically free forever once you've installed them.
[212.0s] So between those two things, those free things, you can really reduce the amount of tokens the years as well.
[219.0s] What else can we do here? The other thing that I like to do is when I'm testing out all of the models, with the same agent, the builds out the HNOS, with all the creations I just put them on Goldiebench so that I can document the process here.
[234.0s] There's another good way to generate leads, because people are searching for interesting stuff, people aren't in no what you've built with it.
[239.0s] People are interested in your agent or SOT if you actually share what you're working on and everything else, especially if you're working on some new unique stuff.
[246.0s] So you can actually give your agent or S access to NetFies well, and then it can document the whole process, the whole building of the agent or S on what you've done.
[256.0s] Kind of like a journal, but it's not you update, and it's not you writing this agent, which is really cool to us.
[262.0s] So that's basically it. That is the whole system. How to build an agent or for an assistant?
[267.0s] How to customize it as much as you want? How to add a memory inside there? Which agent do you use to build it?
[272.0s] How to reduce the amount of tokens you've won. Also, how to create a system that actually works rather than just breaks?
[280.0s] Also, how to use it in a simplified way using our templates. Now if you want to get my full agent or for an assistant, you can get that inside there,
[287.0s] or for boarding inside the Aginal S system section right here. So we update the stately and get the set file for it and you can get the last, you see the last update date here.
[295.0s] You can get a full guide and how to use it, and then every time we build something new.