Today, I discovered a new agentic framework called Goose or codename Goose, developed by none other than Jack Dorsey, the founder of Twitter.
I’d like to share my experience using it. Goose is an excellent assistant that can assist us in various ways, including coding and executing commands. What sets it apart is its free-to-use nature, as it’s open-source. Additionally, you can configure an LLM like Google Gemini Flash Model, which is also free. This means you have an agent at your disposal for free, and it’s truly remarkable.
However, there’s one downside: it can be a bit slow. In the future, they may improve it by adding more models, but it might work faster if used with Anthropic, which I haven’t tried yet.
Now, let’s compare Goose to chatbots like ChatGPT, Clause, Grok, and so on. Goose is far more powerful. It not only answers your queries but also performs tasks as an agent.
I chose to install the CLI (Command Line Interface) version instead of the desktop version. The reason is that Goose requests keychain access, and to access that, I would need to provide my password. Honestly, I don’t remember it, and I wasn’t keen on giving access to my password. Plus, the CLI feels more like a programmer’s tool. With just a few commands, you can get it up and running.
Please note that I’m using a MacBook, so this guide will be easier for those with MacBooks. If you’re a Windows user, open PowerShell as an administrator and install WSL and the default Ubuntu distribution using the following command:
wsl —installRestart your computer if prompted.
Next, run the following command:
curl -fsSL https://github.com/block/goose/releases/download/stable/download_cli.sh | bash
I’m not entirely certain if this will work, as I don’t have a Windows computer. You may even check their official documentation in https://block.github.io/goose/ . If you’re on a Mac, follow these steps:
Open Terminal.
Type the following command and press Enter:
curl -fsSL https://github.com/block/goose/releases/download/stable/download_cli.sh | bash
The command above will automatically install the latest and stable version of Goose on your local machine.
It will then display a welcome message and ask, “Which provider should we use?”
You can select Google Gemini, and then it will prompt you to create a key. You can do this at https://aistudio.google.com/apikey. Click on “Create API key,” enter a project name, and then click “Create API key.” Copy and save the key somewhere safe, and then paste it into the terminal as shown in the screenshot below.
Note that under ‘Enter a model from that provider:’, you may find a default gemini flash model, change that by typing gemini-2.0-flash-exp.
Then it will show -> Configuration saved successfully
Then type goose session . This will show the below:
Goose is running! Enter your instructions, or try asking what goose can do.
Start chatting with goose by asking anything. To close the session, just type exit and you will be out of the session. Ensure that you go to configuration by typing goose configure and then select Add extension by using down arrow keys and then hit return or enter and select Built in extension, then select any extension you like however I think it is better to add Developer Tools which gives it code editing and shell access, you may also add computer controller and memory. I asked it to build a snake game and it did it right away but it took some time as it had to install some packages and dependencies however it was able to the write codes and run the shell command to run the game, a screenshot of which is given below.
I observed two problems: the snake object was moving too quickly, and the high score wasn’t being updated. I requested a solution for these issues. The speed of the snake was reduced while it moved and increased when it consumed its food, as depicted in the image where the white dot symbolizes the snake and the green dot represents its food. The high score also refreshed itself whenever a player achieved a new record. It functioned flawlessly! Each time I launched the game, it displayed the same high score, which was fantastic.
Next, I decided to give it a more complex task. I asked it to find a specific file on my computer, and I’ve included a screenshot of the result for your reference. As you can see, it not only found the file but also provided the path where it’s saved.
Lastly, I requested the deletion of a specific file. I was skeptical that it might not be able to perform this task, but it surprisingly did. I’ve attached a screenshot of the same for your reference.
Now, I can confidently say that we are living in the era of AI agents. If you need any help or assistance, do let me know in comments section.
