CodeLLM - The New Agentic AI Code Editor With Access To Top Models

--- type: transcript date: 2025-02-17 youtube_id: ZMn8ff1vrUE --- # Transcript: CodeLLM - The New Agentic AI Code Editor With Access To Top Models in this video I'm going to be showing you code llm which is an AI code editor and it allows you to access models like Sonet 3.5 deep seek R1 01 mini as well as even 01 one thing to know with this after I recorded this video as soon as the open AI 03 mini High model was released they also included it within the editor I'm not going to be showing you that model in this video but just know that you will be able to access the full O3 mini High model within the editor this is what the platform looks like so just to demonstrate some of the abilities first up is autocomplete with autocomplete there's two main ways that I usually use autocomplete it's very similar to something like GitHub co-pilot if you've used that before this will feel very familiar one is first writing a comment of what you'd like the autocomplete to do if I say given two sorted arrays of nums one and nums two of size M and N respectively return the median of the two sorted arrays in JavaScript if I go to a new line here we have a solution to the problem what's neat with this is you can really just describe with natural language what you want to have happened here if I just say CL it knows that this is a console log it has the context of what we're working in it will also even give you the output of this invocation of that method that's one particular way that you could use autocomplete now the other way with autocomplete the way that it works is if I go and I say something like player and within this object let's say I want a number of different aspects about the player now if I specify what weapons it knows that this is potentially going to be an array and then within that array it can be an object and we'll have a sword and we can just sort of go through there's a bow there's a dagger you can specify the damage and you can just sort of continue and go through your workflow like that this is the autocomplete feature now another way that you can use this is if I just select everything and I command I it will allow me to change whatever I have selected if I just say I want to change this to make it more complicated and have a better structure and after I submit that what it will return is a diff of what I asked for here's our updated player and just like I asked it's a fair bit more complicated so we have our profile we have our stats we also have combat and within that we have our weapons we have inventory and in that case we used Sonet 3.5 what you can also do is if I command I again you also have access to all of these different models let's say it's a particularly tough problem you can go ahead and send this directly to whether it's say deep seek R1 or 01 now that we have our player and more or less a basic structure if I go ahead and select this again let's say I want to use a really powerful model for the next generation and I actually want to have it generate the game cor us I'm going to say based on this character I want to generate a RPG game include everything I need within the JS to get started I'll go ahead and I'll submit that now if you've ever used an o1 model or R1 the one thing that you'll know with those models is they go through a process of thinking and reasoning depending on the query that you ask it definitely will take a little bit more time here we see our diff if I just go ahead and accept that within here we have our player we have our enemies we we have this attack enemy function we have enemy attacks player we have level up we have start game and we have all of these different aspects that we can continually build upon and one of the big differences with code llm where it is particularly unique when you compare it to some of these AI code editors is it gives you a lot of access to 01 what you can also do is generate code within this panel here let's say I ask a one and I say I want to build Tetris completely from scratch I want the HTML CSS and JavaScript to be with within one HTML file now if you haven't used o1 it's incredibly powerful questions like this often times I find are pretty fun questions to ask some of the latest coding models within this panel you have a few different options so you can copy the code you can insert it or you can now apply similar to something like cursor if I just were to insert this for instance we can see everything that we have here we have quite a bit of code here so it generated a few hundred lines for us and let's just just take a look at what it generated for us so if I go ahead and I click Start game here we have Tetris I'm able to move with my arrow keys but one thing that I noticed is as you see on the right hand side there is this Blank Spot so what I could do is let's say I don't want to wait as long for 01 I could pivot back to Sonet 3.5 and I could say I noticed on the right hand side that there were two empty columns help me Resolve that issue I'll go ahead and I'll send that through now the great thing with Sonet is it is considerably faster within here we see that the issue occurs because we have 10 columns with 20 pixel blocks we need the canvas width to be exactly 200 instead of 240 just to demonstrate this if I go and find the canvas element within the HTML so there's our canvas now if I just go and click this apply button what it's going to do is it's going to take that file it's going to take that change and it's just going to merge it with it here here we see that new change we have width of 200 and we have height of 400 if I just save this out and I go back to our game here there we go it immediately made it smaller the one thing in particular where this stands out from some of the other solutions that are out there is well a tool like this doesn't have access to tools like multifile edit or some of the other features that you see within some of the coding editors what what you do have access to are these Frontier models and you have very generous rate limits I'm going to say I want to continue building this to make it look professional and have a linear gradient background a header and a footer and call it developers digest Tetris we have our response back let's go ahead and we'll insert this now we have our file up to 376 lines of code let's save that out let's see what it looks like and we have developers digest Tetris it's a little bit better but it's still not great right let's say we didn't quite like that and we want to take inspiration from the tetris.com site what do you can do is if I were to take a photo of this for instance and I go back to code llm what you'll be able to do is you can actually add and upload files within yours I'm going to say I want my version of Tetris to be inspired by this website have a lot more color and make it look much more professional here we go we can insert this and again we have quite a bit of code here so we're up to 400 lines of code if I go back to our Tetris game it looks a bit better now mind you this is on the left hand side we would have to ask to fix that up or just fix it up our since I didn't quite like the response from 01 let's try son it again let's just say I want to improve this and have the tetris full screen as you start to use some of these state-of-the-art models you'll notice that some models do have particular strengths Sonet is really good at being able to get quick responses as well as overall pretty good code I generally say that most people probably do use Sonet day-to-day and so that's the nice thing with this is if you're really looking to use the absolute most powerful models for coding this is really the only editor that I'm familiar of that allows you to access with very liberal rate limits models like 01 R1 as well as sonit 3.5 the number of other Solutions out there they do give you a credit based system where it's a number of tokens per month to be able to access models like Sonic 3.5 and o1 and 01 mini these are not cheap models one thing that you'll notice with code llm is very quickly after new models come out it's within short order that they roll them within their platform and this is regardless of price or provider if you want to access 01 or deep seek or let's say we have a stal a model that comes out from anthropic it will without a doubt be included within this as well now another thing to note with the platform the access code llm it does cost $10 per month and within the subscription you also have access to their web interface and within here there are a ton of web app features I'm not going to dive too much into chat LM I do have a video on this as well on my channel if you're interested in checking it out but just to give you an idea within here you'll be able to access models like gb40 01 mini 01 R1 and the list really goes on they even have Gro they have deep seek B3 they have the Llama series of models and there are really just a ton of features built into this now the other thing with this you can easily select the context of the different items that you like you can also individually select if you want to reference this piece of JavaScript that we wrote and the HTML you can pass in the context just like that and you can also just chat with it as well I could say something like what does the reset button do explain how it works for instance I'll go ahead and I'll that and within here it will give us a description of what it's doing here's our response looking at the code the reset button functionality works as follows the reset button has an event listener that triggers these actions when clicked we can see all of the different steps here it even gave comments describing what each aspect is doing and it gives a further description as we go on that can be a potential use case if there's something that is maybe a pretty hairy bug within your code you can just pivot to 01 know that you're getting the state-of-the-art model you don't have to copy and paste things within something like chat GPT you'll be able to quickly just reference that piece of context within your code Base by selecting those relevant pieces and including them just like that so next within this list we see that we do have something called code llm what this will do is if I ask a particular query if I ask something like in my HTML I want to improve this make it look more professional and add in a few navigation items and a professional looking footer well go ahead and I'll send that through now as soon as I sent that in you see that it grabbed the relevant context of the example. that we have within our workspace what that's done is it's used some form of model router to effectively triage that request so if you haven't used model routers before how they work is it will take a very quick look at your request and it will determine okay should this use son it and you don't have to worry about pivoting back and forth it will go ahead and attempt to use the best model for the job depending on the request so and the other thing that I want to mention to close out the video is at time of recording they're actually offering an unlimited quota as an introductory offer for sometime to be able to access 011 mini deep seek R1 quen Sonet 3.5 and I wouldn't be surprised as soon as we start to see other models whether it's from anthropic or Google models that are particularly good at code we'll also see them rolled within a platform like this but otherwise that's pretty much it for this video if you found this video useful please like comment share and subscribe otherwise till the next one