Free Cursor Alternative? Trae AI Just Got Way Better

If you are into Vibe coding, I think you are going to really like this one. So we're going to be looking at Trey AI, which is a free AI coding assistant. In fact, it's a full fledged coding IDE that is built on top of VS code and it's also a sponsor of today's video. Now, they recently had a huge upgrade. So now you can customize your coding agents for specific tasks with specific set of tools. And with MCPS going viral everywhere, they also added support for mcps. So in general, it's a really good free alternative to some of the other coding ides out there. In this video, I'm going to walk you through the new features in Trey AI and how I plan to use them. Trey AI probably has one of the cleanest UI of all the coding IDs out there. And their latest features definitely makes it a really good option for wipe coding or coding in general. So let's get started. Trey AI has completely redesigned their UI and have added some really cool new features. These includes new agents that you can customize and now they have support for mcps, which makes it extremely powerful. They also now support customers custom rules for your projects and they also added the ability to do web search. Now, Trei has all the basic features of any other coding ide, but I think the custom agents is a very special feature. I'm going to show you a couple of examples later in the video. Now, Trey AI has a number of different models that are supported, including the Cloud 3.5 sonnet and the latest Gemini 2.5 Pro. So here's the list of model that are available for free at the moment. But if you want to use a custom model from a provider, you can actually do that. So for example, if you want a model from OpenRouter, you can select that. You can select the model that you want. Let's say we want to use something like OpenAI01 and then you will just need to provide your API key and you will be able to use that model right away. The feature that I personally like is this new custom agent. So you can build different agents and you can customize them. So let's say we create a new agent. I'm going to call it Web Search. Now you can customize your agent behavior by providing a custom prompt and then you can select the different tools the agent has access to. So at the moment you can see it has access to preview web search terminal file system. But let's say instead of this web search, we want to create an MCP that is going to be able to do web search for us. So I'm going to remove this file crawl mcp. Now, if you go to the list of mcps, there are some preconfigured mcps. So you can choose anything from this list. It basically works as a registry of mcps and the UI is pretty nice. Let's say I want to use tavli, which is an API that enables you to do web search. So the way you add an MCP is this. Here are the configurations that you will need to add. Now for the MCPs that are listed within the tray AI, they also have a link to a web page where you can actually go and read the documentation. So it opens a preview of the web page right here in your ide. So you actually don't really need to go to a web browser if you are just looking at the documentation. So let's say in this case we just need to copy this configuration. So I'm going to copy it, I'm going to come back here, we're going to paste it and all we need to do is to provide our API key from tell the website. Okay, so I'm going to paste my API key. I am going to be removing this right after recording this video. And now it says automatically added to builder with mcp. I'm going to say I understand. And now we can go back to our custom agent and we can enable that. We're going to click on create and we will use it right away. Okay, so I'm going to create a new chat. And now within my agents I have the builder which is the default agent for coding task. Then there is this builder with MCPS that is going to have access to all the MCPs. But now I created this web search agent which only has access to the table MCP and this really lets you customize your projects. You can use very specific agent for very specific set of tasks that has access to a specific set of MCPs. I asked you to look up the latest Gemini SDK. So first it looked at the current workspace, but then it decided to use that web search MCP that we created. Now it's going to ask for permission before executing a call to the mcp, which is critical, especially you are giving access to your system in some cases. Or you can just enable Autorun, which I'll not recommend doing at all. But let's run this. Okay, so based on the web search it says the latest version. So now Google uses this unified SDK, is able to figure that out for some reason. It's also listing the Gemini 1.5 Pro, which is like an ancient model at this point. But that was a quick example of how MCPS can work within Trey AI. But let me show you an example of a project that I would personally use Tray AI for. So let's say we're going to start a new chat right now I'm going to just select Builder this. That's basically the agent implementation. Now you can select any coding model that you want. I will go with Gemini 2.5 Pro. Sometimes you might run into rate limits. In that case, either select a different model or you could potentially use a model from OpenRouter. So I created a requirements doc for a project which is going to be a text to image generator. This is going to be a client side web application that will allow the user to generate images based on text prompts using the Google's Gemini API. We specifically want to use the Flash 2.0 model because that has native image generation capabilities. Okay, so the user is going to provide a prompt based on that prompt. We're going to generate an image using the REST API from Gemini and we'll display the image back. Then has to be support for image regeneration, downloading and also we'll just show an animation while we are waiting for the image generation. The API is supposed to look like and then the file structure the project is supposed to look like. Okay, so in order to use this as a context, all we need to do is just ask the builder to implement the project based on the requirements. Now in order to add the requirements, we're going to be using the pound sign. You can provide a code, file, folder, workspace, docs, or you can just point it to do web search. In this case we have a file, so we're going to use the file and then we can just select this. Okay, we'll send this in. Okay, so it has implemented the project and here it's asking about the API key. So we need to provide the API key. I also looked at the API endpoint that it's using. That seems to be correct. So let's see whether this works or not. For my personal projects, I have moved away from using any of the SDKs. If a service provide a REST API endpoint, I usually use those now. And as you can see, that's the case here with the Gemini API as well. It makes maintainability of the code very easy. Now since it's an agent, it can actually run the code and look at the output. So, so we're Going to run this right here in the terminal. You will see it's running the Python code. It also lets you have a preview of the web page. So if I click on this, it's going to show me a preview of how the web page is supposed to look like. Now, this is supposed to be fully functional, so let's see if it's actually able to generate images. We're going to ask it to create an image of a llama playing soccer. Now, instead of the image, we got an error, so let's figure out what happened. Okay, so here's the actual API call. I'm going to just copy that documentation. Now, it also uses multimodal inputs, so rather than describing the problem, we're going to just take a screenshot, send it in, and I'm going to also tell it what exactly is needed in order to use that REST API endpoint. Okay, so it made some updates and let's see if it actually is going to work. Okay, so we're going to try the same prompt again. Let me just make it bigger so that we can actually see what's going on. Let's click on generate and yeah, we have an image of a llama that is playing soccer. Let's see if the regenerate functionality works. Yeah, that seems to be working as well. And let's check download. So it does seem to be able to download files as well. Now, overall, it seems to have all the functionalities of the coding AI assistance at the moment. So for example, you can just select something like this and ask ask it to explain it. The coding capabilities are dependent on the model that you choose. Gemini 2.5 Pro or Cloud 3.5 Sonnet are really good at coding. They even have support for the latest GPT 4.1. And again, all of these are available for free, but you might run into some rate limits. Also, you can edit the code right here. So maybe I can ask it to add comments. The UI seems to be very clean compared to some of the other coding IDEs that I have seen, because here you can actually see the diff. So let's say we accept it and it added those comments. Now, I haven't tested this for large projects yet, but for smaller projects seems to be working fine at the moment. Tray AI is free to use, so one of the biggest concern is going to be privacy and security. And this is something which they have directly addressed on their website. So I'm going to read directly from here. Take it for what it's worth. So they say local data storage. Code base files are stored locally on your devices for indexing. Files may be temporarily uploaded for embedding. After processing, all plain text is deleted. Now secure data access are strict access control and encrypted transmissions are enforced to prevent unauthorized access and reduce exposure to security risks. And then regional deployment. User data and infrastructures are deployed based on account location stored in the us, Singapore and Malaysia with isolation in place to meet local data regulations. Now overall, I think it's a pretty powerful coding ide. Especially it has these neat features of custom agents. I'd highly recommend to check it out and do let me know how it compares with other coding IDEs like cursor and Windsurf. Anyways, hope you found this video useful. Thanks for watching and as always, see you in the next one.