AgentsInTheCloud - Random Stuff I Like

Hey, you know how models these days just like to use bash for everything and it's actually kind of hard to see what they're doing.

Like like this guy over here, you'll see sort of like robot barf like this.

And we have a feature for that.

And like we render it like this.

So that means that we reformat the bash, we lay it out, we syntax, highlight it and then we also make sure that we detect things.

For instance like this here is actually inline Python.

So that gets syntax highlighted as Python.

Then it actually also sees that this is inline JavaScript again inside of the python.

And I love that little feature because I actually like to keep an eye from time to time on what the agent just when I'm trying new models and just to sort of, you know, not to lose too much touch.

And then there's another feature I like which is for when the tool call is still in progress.

I really like it to get an idea of what it's working on.

I don't like to sort of be staring at this working for 10 seconds.

So here you can see it's actually giving us a live terminal view of the command as it executes.

So that I know what's going on.

It's nice when you have build systems that have really nice ANSI output, stuff like that.

there's actually another feature since we're in the final video and I can just show off whatever I want.

let me turn this into a 200 thing.

And you can see that the bash commands actually have a stop button here.

And I find that very often models will emit very long waits.

And if I'm in a rush I often decide to stop it.

And when you stop it you get actually a dialog where you can tell the model, your desired result for this tool call.

No, stop it.

But do it again with a five second one.

And now you can see in the tool call that the result that it gave the message that I had.

So I use that a lot when you like I do releases and launches from a workspace like this.

And often the it needs to wait on a Docker push or something and it just loops with 60 seconds, waits in a for loop.

And then when you see that it's done, you can actually just abort the weight like that.

let's see, do I have any other last minute things I want to show?

I think in none of the videos I showed this park thing, but you can do this park, you know how Sometimes, you know, sometimes you fire off like dozens of workspaces and you know, you're hitting off features of like small stuff one by one and you're in a great flow.

And then sometimes you get an answer back from the agent and it's like, and you realize like, this one I need to think about.

Like this is not easy.

This is going to require a bunch of focus and, and sometimes you can do that right away, but other times, like I want to give it that focus, but not right now.

And I also don't want the workspace to continue to remind me that it exists.

So then I click park and it basically parks it in here.

Sort of like out of sight, out of mind for a while.

It also stops the container then, so you get your resources back.

let's see, do we have anything else for the final video?

I really like this usage thing.

It's a ring.

let me see.

So for instance, this ring here means that on my anthropic 5 hour limit I am pretty good because the, the time is basically.

No, the time is basically here, but my usage is here.

So the green means that you're doing well, you're like, you know, like you're behind pace.

Whereas on the seven day cloud I'm a little, you know, it just reset today probably.

and I'm a little bit ahead of pace.

And on my codec subscription I am also a little bit ahead of pace, but not by that much.

The icon that you see in the top left here is the limit for the provider that you like, the limit that you're most likely to hit first of any provider that you have used in the last hour to do any inference.

then why not do a few more features?

when you go to new workspace you have your settings here and in the Docker one, I don't think we talked about that, but you can do a custom Docker file here.

So if you are working on a project that requires some system install, maybe like the Rust compiler or something, you know, something that is kind of big and that you don't want to pay for every time, you just write your own Docker file in here, you have to derive from the default one, you can add whatever you want so that when you add a new workspace it is quick and ready to go.

All right, that's it.

Have fun.