Getting started with Antigravity CLI

So Google did it again. At Google I/O they announced the migration of Gemini CLI to Antigravity CLI.

With that in mind, I put together this very quick run through so you can have a play and see what you think.

Application Download

Downloading the application is pretty straightforward, Google have created a landing page at https://antigravity.google/download.
Here you get the choice of macOS, Windows and Linux with support for Apple Silicon, Arm and x86. All of the above have pretty minimal requirements for hardware due to this being a remote cloud model solution.

From what I can tell, the Antigravity CLI appears to be a Go binary? So no more Python required. I personally prefer a binary distribution as its easier to bundle into a container or Nix package. Not having to deal with pesky pinned versions or other bill of material issues is also welcomed.

It also looks like someone did some engineering as the install script does some detective work to ascertain the platform. It would have been nice to have these artefacts continue to be open source, but it looks like that is not the case with Antigravity CLI (at least for this release).

Pop into your local terminal and enter the magic install command:

curl -fsSL https://antigravity.google/cli/install.sh | bash

Once the installation is completed successfully, you have yourself a binary on your host called "Agy".

Antigravity Configuration

With the application successfully installed, Antigravity or Agy will need a bit of configuration. First up is authentication, where you need to decide between Google OAuth or a Google Cloud Project. Once you select, click on the provided link to authenticate and get your personal authorisation code.

For the purposes of experimentation I selected a gmail account to authenticate. If you wish to use a Cloud Project, do take care to ensure you restrict access to avoid any unpleasantness.
With login authentication out of the way, you can then make the interface match your aesthetic of choosing. I wanted to go for solarized dark, but the tokyo night colour scheme was totally my vibe.

You will also need to agree to the Terms of Service and Data Use. Helpfully you can opt out of Google data collection and using your interactions.

Finally you need to provider Agy with permission to access your workspace. Agy will attempt to write to your computer, so its worth setting up a location first, before doing any tasks.

Antigravity First Steps

After all of that you can now put Agy to work. Out of the box, Gemini 3.5 Flash is available to you in this version. If you don't plan to do anything fancy, you can immediately get to work and start to use the application in a similar way to the Gemini web page.

First thing you will probably want to do is checkout the help option. Antigravity uses the /command method. So typing /help will present some help information. You will notice just typing the / command presents a list of options. It doesn't look like you can type /help [command] which is a shame.
The second thing I tried was using the /models command to give a list of available models.

In terms of models, the following were available with my installation:

> Gemini 3.5 Flash (High)      (current)
  Gemini 3.5 Flash (Medium)
  Gemini 3.1 Pro (High)
  Gemini 3.1 Pro (Low)
  Claude Sonnet 4.6 (Thinking)
  Claude Opus 4.6 (Thinking)
  GPT-OSS 120B (Medium)

Clicking on the model option immediately switches to use that model. It also has a nice feedback at the bottom of the viewport to indicate both the model and level.

Another thing to pay close attention to, is the available quota for the model in use. Use the /usage command to see the available quota for your current setup:

In terms of user feedback, usage quota information should be available on the /model configuration. I would also like to see an indicator on the main user interface to avoid confusion.

Antigravity Text Generation

Now we have looked at a couple of settings, it time to have a play. So lets generate some text. So according to Google I/O 2026, Gemini 3.5 is awesome. Lets see if Antigravity can evidence and articulate this better than Sundar :-)

Watching Google I/O Sundar says Gemini 3.5 is awesome. 
I am not sure why, can you explain how this model compares to previous versions to help me understand this better?

Based on the above prompt, Antigravity initiated a web search in which it thought for 4 seconds and used 522 tokens. A summary of the response is shown below:

In short, Gemini 3.5 isn't just slightly smarter or larger; it is designed to act rather than just answer. By combining massive output limits, extreme speed, and native reasoning (which keeps track of its own thoughts over a long conversation), it enables agentic workflows that were previously impossible or highly prone to failing mid-task.

Ok, not bad. We need to give Sundar some credit here, his explanation was way better (and funnier) than Gemini!

Antigravity Coding

The next thing to try was the coding element. Personally I prefer to not code this way, but I totally get the appeal. Let pick something other than a todo list!
I would like to build a terminal application that can read rss feeds. I will provide a URL and then the headlines are displayed for the site if it supports RSS.

From the initial prompt, Agy comes back with a implementation plan. The implementation sets out what Agy understands from your requirements and how it will go about implementing this plan.
Agy also looks to utilise tools required for the job (and if necessary requesting user permission). During the create process, Agy will prompt you for any interaction required e.g. permission to create or update or delete.

I did notice that tab autocomplete didn't work for me with the bash commands. Also I couldn't get the Python terminal application created to run directly in Antigravity.

At the end of the process a Python rss reader was born. Great job Agy, its pretty cool application.

Summary

Agentic coding harnesses are all the rage and with good reason. They offer a significant amount of flexibility and simplify common tasks. Adding Agy into my workflow I can definitely see the appeal of using a tool like this.

The migration path is a little bumpy. However, the inclusion of an installer, is a great benefit. Additionally previous releases are also available, so users have a great choice available to them. Personally I like when software can tell me which version is available in the command line options - surprised to see this missing. Also maybe include an uninstaller in the next version? Pet peeve with a lot of these coding agents, is they take liberties with your hard disk, installing files all over the place. For the installer crowd, uninstall destructions below:

1. Remove the local content

rm "$HOME/.local/bin/agy"
rm -rf "$HOME/.cache/antigravity"

1. Edit your environment script to remove the Antigravity CLI reference

# Added by Antigravity CLI installer
export PATH="/home/labdemo/.local/bin:$PATH"

Overall I think the Antigravity team did a tremendous job on the CLI. Its pretty straightforward to get going and it feels familiar if you are used to other coding harness software. For existing Gemini CLI users, get started on those migrations as support is going away pretty quickly.

If you are team Google, this is definitely a great no hassle option. Surprised it doesn't include Gemma, but assume that will be incorporated in later versions.