Transcription
A while back, we explored Trey AI, an AI-powered, fully free open-source IDE that was developed by ByteDance. It gives you access to state-of-the-art models like Claude for Sonnet as well as Opus. You get access to OpenAI models as well and many others, all without spending a single penny. It's surprisingly powerful and stands out as a strong free alternative to tools like Cursor as well as Windsurf, offering features like real-time code generation as well as debugging and many other features that would assist you with your development workflow.
And just recently, they introduced a new open-source tool called Tray Agent, an interactive CLI-based AI agent that can execute complex workflows from simple English prompts. It supports many models and it allows you to automate development tasks like running shell commands and managing project operations, all through natural language within your terminal. And what's also great is that ByteDance is putting some serious heat on Cursor with this open release of Tray Agent, which recently topped the Sway Bench verified leaderboard with an impressive 75.20 percentage score.
But in short, Tray Agent is an open-source, MIT-licensed AI agent for general-purpose software engineering tasks. It's built with Python, and it features a CLI that can execute complex workflows using tools and large language models. It's similar to Gemini CLI as well as Cloud Code, but with different feature sets. You have a couple of different features listed over here, like the lake view, which provides a short and concise summarization of agent steps, different large language model support, rich tools, MCPS, and interactive mode, and much more. It's essentially a modular, research-friendly architecture, and it's ideal for customization, experimentation, and for various other use cases. It's currently in alpha, so they're still going to keep on working on this as time goes.
Before we get started, allow me to introduce today's video sponsor, Outskill. Remember when Steve Jobs unveiled the first iPhone in 2007? That moment changed the world. But what took Apple 15 years to make normal, AI is doing it within 15 months. We've officially entered the AI adoption explosion, and Outskill is making sure you don't get left behind. Outskill is hosting a free 3-day AI Mastermind, a 16-hour live sprint that'll take you from AI curious to AI confident. Learn how to build agents, automate your work, make stunning videos, and even launch your own AI startup. All taught by experts from Microsoft, Amazon, and Samar AI. Normally, this costs $895, but I've teamed up with Outskill to get you one of the first thousand free seats with no strings attached. Plus, you'll get over $5,000 worth of bonuses, like a 3,000-plus prompt bible or a money-making mind map, as well as a personal AI toolkit. Grab your free seat using the link in the description below and the pinned comment. Spots are filling fast, and the kickoff call is this Friday at 10:00 a.m. EST. I'm personally making my whole team join this. And if you're in tech, marketing, freelancing, or just want to future-proof your career, this is a no-brainer. So, with that thought, click the link in the description below, join the WhatsApp group for updates, and I'll see you inside the mastermind.
So, to get started, it's really simple. You're just going to need to make sure you have the prerequisites fulfilled. You're going to need to make sure that you have Python installed. This is 3.12 plus. Make sure you have Git installed to clone the repository. And then have an API provider. In this case, I'm going to be using Open Router, as they provide a free API key that you can use. It's going to be rate-limited, but it still will help you get the job done. But you also have an option to use OpenAI as well as Anthropic for whatever use case you want within the CLI.
But to get started, we can head all the way to the top of this repo, which I'll leave a link to in the description below. And what we want to do is open up our command prompt. And once you have it opened up, you want to go ahead and you want to copy this GitHub repository link. Then you can scroll back down to the installation guide. And then you want to type in `git clone` and then paste in the link for this repo, which will clone it onto your computer. Once that is done, we want to go ahead and go into the `tray-agent` directory. And that is by typing in `cd tray-agent`, which is by simply pasting this in. And you can see that we're now within this `tray-agent` directory. And then to install all the dependencies, you can use the `uv sync` command, and it will help you install everything necessary.
Then all you got to do is just set your API key using the `export` command, or you can go right into the config file. And guys, you can also configure all the different like configurations, like the max token temperature, within this JSON file as well. And guys, all you got to do now to start it up is simply run the `tray-cli` command. And this is where it's going to start up easily within our terminal. And you can see that Tray Agent interactive model is now running. And we can start using all these different tasks.
Now, there's a couple things that we can do to execute a task. You can use the basic task execution to create anything by running the `tray-cli` command and then, uh, typing in `run` and then your prompt. Then with specific providers and models, you can put in your prompt and then select the provider that you want to use as well as the model of your choice. Now, if you're using Open Router, you can do that as well by selecting the provider, which is Open Router, and then the model of your choice. And then, if you're working with a custom directory, you can set the path that you want to work within. And there's a couple other things like debugging as well as patches.
Now, there's a couple of modes that you can run, like the `status` to see more about your agent, `help` for available commands, `clear`, `exit`, all the things that are typical within any CLI tool. But what I want to showcase now is the available tools. There's so many different tools associated with Trey AI's agent. You have the ability to use the `edit` tool, which can create, edit, view, and manipulate files. You have the ability to execute shell commands as well as scripts, sequential thinking, as well as signaling a task completion.
And now all you got to do to run anything is just run the `tray-cli run` command and then putting in your prompt in natural language. And you can see that I set my working directory as my desktop. The trajectory will be saved within the JSON file. And then it will execute the task. And you can see now it has created this, uh, "create a hello world" Python script. And like I said, it has now saved the trajectory file of the run session, and you can see that it was able to create this "hello world" Python script, and it gives you the logs of it. Now, you may ask, what is this going to be used for? Well, you can use this for debugging as well as just understanding what is happening in the background as you're running, as well as executing things with AI, so that you can get more familiarized with whatever the AI is generating. And this way, you can use this to error-handle all the generations.
Now, let's have it work on something a bit more intricate. Not super intricate, but a simple Mind Super game. And the reason why is because I want to showcase the toolset that Trey actually provides. So, in this case, we're going to have it send in this prompt to create the Mind Super game. And we're going to specify the working path. And there we go. Just like that, we have a Mind Super game that it was able to actually develop pretty quickly. You have three different modes: easy, medium, and hard. And you can see that it was able to fully develop it within a couple seconds with different sorts of small features like animations as well as modes.
Now, I know a lot of you guys in the comments are going to be asking me, what's the difference between Tray CLI versus Cloud Code or Gemini CLI as well as even Open Code? Well, the main difference is that Tray Agent as well as the CLI itself can actually execute commands, edit files, and automate workflows using modular tools like bash. Now, if you take a look at Cloud Code, it's not actually able to use these `str`, `replace`, `space`, `edit` commands because it is something that operates purely as a chat assistant that generates the code, but it's not able to actually run it. Whereas Tray Agent performs real actions in your system and it's able to support multi-step reasoning through these structured agents.
Now, I'm not saying it's way better than Cloud Code because, in my opinion, I do prefer Cloud Code due to its GUI as well as its additional toolset. But the fact that you're going to be able to execute commands based within your terminal is definitely a great feature to work with.
In conclusion, guys, Tray Agent is definitely a nice new addition to the open-source space, and I really love the CLI agent that they have developed. Hopefully, they can keep on adding more large language models to be supported as well as more features to make it more extendable. Now, currently, the UI of the CLI is decent, but the GUI needs a lot of work, and I hopefully they can still improve on that as time goes and add more features. But that's basically it, guys. I hope you enjoyed today's video and got some sort of value. Let me know what you guys think about Tray CLI. Curious to see what you guys actually think about it.
But with that thought, guys, thank you guys so much for watching. I hope you enjoyed today's video. Make sure you take a look at the newsletter, follow me on the second channel as well as the Discord, and then make sure you follow me on Twitter as well as subscribing to the YouTube channel by turning on the notification bell. But with that thought, guys, thank you guys so much for watching. Have an amazing day. Spread positivity, and I'll see you guys fairly shortly. He suffers.