Name: OpenHands
Uploaded: 2024-12-07T17:00:52.609Z
Duration: 2 min 33 s
Description: An open-source platform for AI-powered software development agents capable of performing tasks akin to human developers.

openhands

An open-source platform for AI-powered software development agents capable of performing tasks akin to human developers.

OpenHands

We officially now have the best and new AI software engineer that is completely open-source and even capable of beating Claude 3.5 Sonet. It's called OpenHands, which is a framework that we had covered multiple times on this channel. 

Now recently we had made a video on Open Hands covering its new major updates such as its development framework extension with the new UI as well as a few minor updates. But the depths behind the project are back again with a new extension to the framework called Kodak 2.1, which now achieves state-of-the-art results on the sued bench benchmark test. 

It currently achieves a 53% resolution rate on the sued bench, which surpasses even claw 3.5 Sonic, being the best coding and software development AI agent. Before we even dive into this update, allow me to reintroduce what OpenHands is truly about. 

Open Hands is a platform that you can access today where you can use it to deploy AI agents to tackle software development tasks. Open Hands agents can do anything a human developer can, where you can modify code, run commands, browse the web, call APIs, and yes, even copy code snippets from Stack Overflow. 

It's quite powerful, and now having it connected with the Kodak 2.1 framework, it's even better. Now before we get started, I definitely recommend that you take a look at the Patreon page so that you can access the new subscriptions that will be releasing this week. 

So what is the Kodak 2.1 agent that is now deployable within Open Hands? Well, it's essentially a new powerful coding-based agent that was powered by the new CLA 3.5 Sonic model that was just recently. 

Launched from Anthropic last month and essentially they have configured this model to tackle task even better now in the Suway benchmark test. This is a benchmark designed to basically evaluate LM's capabilities in addressing real world software engineering challenges. Essentially within this benchmark test they were able to record the Kodak 2.1 agent to significantly surpass many of these other frameworks, even Claude 3.5 Sonet, which is just insane to me. 

The great thing is that you can actually get started with this right away and even install this today. They essentially released this new framework that can handle a wide range of tasks by running Python and Bash commands, all driven by language models. Essentially they based this off their single agent research paper that they released, and this is where they have focused on the ability to configure an agent as well as having it coupled with different sorts of teams of agents to tackle tasks in subtasks. 

They have this paper that discusses this further in detail based off of executable code actions, and this is something that I definitely recommend that you take a look at because it's super interesting. Essentially they have the new agents where it's going to be running Python and Bash commands, and there's three main components to it. You have the ability for function calling, which is going to provide more improved precision and letting the model specify functions directly. 

They also are going to utilize the new CLA 3.5 Sonic model so that it could provide more of a better coding based capability to the agent. 

You lastly have the directory navigation fixes which is going to enhance the agent's ability to navigate through different directories. This is essentially what makes this new code act 2.1 agent surpass many of these other Frameworks. So you must be interested now and you may be wondering how can you get started. Well, it's super easy. You have a quick start menu on their documentation which will showcase how you can easily get started. 

There's a couple system requirements that you will need to fulfill beforehand. This is where you need to make sure that you have Docker version 260. So make sure you install the latest Docker version which I'll leave a link to as well as all the other links that I use in today's video in the description below. After you have that prerequisite fulfilled, you need to make sure that you have Python installed. 

And once that is installed, if you're going to be installing this on Windows, you must use WSL which is a subsystem for Linux. So if you do not have this, make sure you install this onto your computer if you're on Windows. But once you have these prerequisites fulfilled, since I'm on Windows, I'm going to simply open up WSL. 

And once I have WSL opened up, I'm going to have Docker opened up. So make sure you have Docker running in the background before you get started. And once Docker is running, you can then easily go back into the GitHub repository. You can simply copy this command over here and you can simply just paste it within your command prompt. This is going to pull the Docker container and it's going to install all the necessary requirements. 

For this project this will take a couple of seconds but once it's finished I'll be right back. 

And it looks like it is finished installing and you can see that it's now running on our Local Host. 

So go over to this Local Host link which you can access within the GitHub repository and you can see that you can start interacting with open hands. 

So first things first you're going to need to provide your claw 3.5 Sonic API key. 

This is the best way possible to utilize open hands. 

It's the only way you're going to be able to access the code act 2.1 agent not code act the rapper code act agent. 

So simply go ahead and provide your our anthropic API key and then we can proceed forward. 

Once you do provide your API key you can see that the UI for open hands if you have used the before has changed even better. 

You can see that you can now start interacting with this by requesting to build things with this single prompt such as writing a bass script to show the top story on Hacker News. 

You can even open up a repository via GitHub or you can import a project by uploading a zip file. 

Something I forgot to mention is that you can even utilize open-source models such as from a provider like Ama or even LM Studio. 

You can see that there's various providers that you can utilize and something else to note is that you can even utilize the anthropic claw 3.5 hiu model which is the new model that was just recently introduced. 

So if you're interested in that definitely utilize the model of your choice but in my opinion like I said Cloud through pumi Sonet is how you can utilize the new Kodak 2.1 agent. 

Let's requested to do this command to write a bash script. So let's go ahead and send this in. Once you send in a prompt, you can see that it's going to start utilizing the same artifact type of UI to execute your commands. It's going to first develop a plan for you and you're going to be able to utilize this workspace to see all the changes being made. 

So we can see now the green light has been given to our AI agent to start executing the task. So from here, what it's going to do is it's going to start creating the simple plan and all the changes as well as execution are going to be made within your terminal and going to be displayed within your workspace. 

So I'm going to go ahead and let it generate the base of the script and I'll be back once it's finished. And there we go, we have the script ready. In this is where we had requested to display the top stories. We have it displaying all these top stories and it has graphically made it look a little bit better. 

So now what we're going to be doing is going back to the main page and we're going to be requesting it to generate something else. But something I forgot to also mention is that if you are to click on the bottom right three dots, you can connect this to GitHub and upload this to a repository or even download this as a zip. 

But let's go ahead and exit this project and let's request it to generate something else. So what I'm going to do is I'm going to have it design an AI chatbot web app that's similar to ChatGPT. So let's go ahead and send in this prompt and let's see what it's capable of generating.

Be able to click on this browser tab which will show you live updates as to what is being generated. So once it is finished generating I'll be right back. 

So look at this. So I went from developing the code and now we have a fully functional EII chatbot that was developed within a couple of seconds because of the Code Act agent. So now let's take a look and basically explain the agents and the capabilities. 

So you have these different types of acting agents and in this case their main agent is the Kodak agent that Kodak has in not the rapper but Code Act agent. And it's designed to simplify and enhance the actions of LM agents. 

What's going to basically happen is that this Code Act agent is going to consolidate all of these different action items into unified code action spaces. So this will basically make the agent more efficient and easier to use and basically have it so that's easier to send in these different actions. 

It starts off with an observation. It will then send it over to the agent which is going to then have it consolidated. It's going to think about it, plan it with different types of methods and algorithms like Chain of Thought, self-reflection, and improving action from prior observations. 

From here an action is sent which will basically have the Kodak agent have unified action space within a different environment or have it sent within a conversation to the user. And within this loop it's going to be able to work on enhancing and having a more efficient and easier way of generating an answer. 

Now at each interaction you're going to be able to see that the Kodak agent.

Can have a conversation meaning that there's going to be a communication with humans in natural language. So this is where you're going to be able to chat with each step of the generation, which is something that we saw with prior examples. You also have codak, which is going to choose and execute task by running code, and it's going to choose the best approach based on the situation. 

You have it able to be executed with any valid Linux bash command or execute any valid python code, which is basically an interactive python interpreter. This is something that is used to run the python code that is simulated through bash commands for more flexibility on the back end. You also have a plug-in system, and this is where they have basically made the Kodak agent more powerful with the use of different accesses to bash action spaces. 

You also have the ability to integrate various other plugins. You have different actions that you can set in, and you can observe with different things like Lang. You have method a planner agent, which is another type of agent, and this planner agent will basically utilize a specialized prompting strategy to create long-term plans. 

This is for solving the different problems that is sent into the framework itself. Just to emphasize on what sort of LMS you can use, like I mentioned, you can possibly use any LM with this framework. You can have a PLL request with AMA, meaning that you can utilize a local large language model with open hands. 

But that's essentially it for today's video on the new code act 2.1 release from Open Hands. This is definitely a game changer. Please share the AI Agent Store with friends.

It's something that I definitely recommend that you try out cuz at the current moment at the current date this is the best framework that outpaces many of the other things that are out there such as claw 3.5 Sonic for complex software engineering tasks as well as Klein AER and many of the others. 

Now I'll leave all the links that I use in today's video in the description below. Make sure you follow me on the page patreon so that you can access a private Discord. Follow me on Twitter a great way for you to stay up to date with the latest AI news. 

And lastly make sure you guys share the AI Agent Store with friends. But with that thought guys have an amazing day spread positivity and I'll see you guys really really soon. Peace out fellas.

Note, that there may be slight differences in text and video. Watch video for the original content.

OpenHands Video