← Back to Blog

Gemini CLI stopped working on personal accounts. Move to Antigravity CLI

· 12 min read Gemini CLI stopped working on personal accounts. Move to Antigravity CLI

If Gemini CLI accepts your Google sign-in and then refuses to answer, reinstalling it is probably the wrong next step. Google stopped serving personal-account requests through Gemini CLI on June 18, 2026. That includes Google AI Pro and Google AI Ultra subscribers. The replacement for this route is Antigravity CLI, whose terminal command is agy.

We hit this on a Windows 11 workstation with a paid Google AI subscription. Sign-in completed. The next request failed. We then installed Antigravity CLI on that workstation and in a Linux desktop container, signed in, and got a successful reply on both. This article combines those October 26 tests with Google's current documentation.

Our AI Terminal Kickstart scripts now offer Antigravity CLI instead of the old Gemini npm package. It remains optional. Here is what changed, how to move, and what to keep.

The error is about account access

This was the exact message in our test:

This client is no longer supported for Gemini Code Assist for individuals. To continue using Gemini, please migrate to the Antigravity suite of products: https://antigravity.google

The error included UNSUPPORTED_CLIENT. Although the account had a paid Google AI subscription, the response identified its tier as Gemini Code Assist for individuals. Browser sign-in had worked; permission to use that service through this client had ended.

Google's release notes, read October 26, name the three affected tiers: individuals, Google AI Pro, and Google AI Ultra. Its consumer deprecation page confirms that this includes Gemini CLI's personal Google login route.

Installing a newer npm package cannot restore access that the service has withdrawn. Switching from one personal Google account to another affected personal account will not fix that policy either. A repository can keep publishing releases while a particular account route stops working. Our older kickstart explanation confused those two things; we have corrected it.

There are exceptions. Google's transition announcement says paid Gemini API keys remain an access route. Existing Gemini Code Assist Standard and Enterprise licenses were also outside the June consumer cutoff. If that describes your setup, keep the working CLI while you assess your next step.

Those business licenses now have a separate retirement schedule. Google's page, last updated October 22, says existing subscriptions stay supported until their final renewal or contract end date, licenses can still be added until January 31, 2027, and auto-renewal ends on a 2027 date that depends on the plan. Your tech person should check the contract and renewal dates. A personal Google AI subscription and a business Code Assist license are different products.

What Antigravity CLI offers

Antigravity CLI installs as a native binary. You do not need npm to install it. The installer says it self-updates during normal use, and agy update gives you a manual update command. Our tests used version 1.4.7; the installer selects the current release, so your version may be newer.

You can use it interactively to work through a task, or use print mode to return an answer to a script. The help output we checked supports text, JSON, and streaming JSON output. That matters when the next step is another program, such as collecting a draft summary into a review queue.

For a small business, a useful first task might be explaining a website folder or drafting a checklist from approved notes. Keep the task narrow enough to check. A successful installation is not proof that the agent understands your prices, refund terms, or customer promises.

The command also exposes effort and execution controls. In the tested release, --effort accepts low, medium, high, xhigh, and max. The --mode choices are plan, accept-edits, and auto. These are separate decisions: how much reasoning to request, and how the session handles work. Choose deliberately instead of copying the most permissive example you find.

There is a --sandbox option for terminal restrictions. MCP connects the agent to additional tools, and plugins package extra capabilities. Start by checking what is connected with agy mcp list. Adding a connection should be a decision about what the task needs, particularly when that connection reaches business records.

Google's transition announcement confirms support for skills, hooks, and subagents, with Gemini extensions becoming Antigravity plugins. Skills hold reusable task instructions. Hooks attach actions to session events. Subagents let the main agent delegate work. You do not have to configure all three to get value from the CLI.

The CLI reference also lists a /browser command that hands web browsing, page checks and in-browser debugging to a browser subagent, which runs in a separate Chrome profile. Remote control is available too. The tested command includes --remote-control and the remote-control subcommands start, status, and stop. That is useful when you want to supervise a session from elsewhere. It is separate from SSH sign-in, which only needs the normal login link and code.

Google also says there was no one-to-one feature parity at launch. Before moving a recurring job, check its prompts, permissions, connections, and output handling. A script built around Gemini CLI's output should not be treated as compatible just because both programs accept text.

The models we saw

On our account, agy models listed Gemini 4 Argon, Gemini 3.8 Flash, Claude Opus 5.5, Claude Sonnet 5.5, and GPT-OSS 120B. The Gemini and Claude entries offered Low, Medium, and High options; GPT-OSS showed Medium.

That is an observation from our account on October 26, not a promise about every plan. Run agy models after signing in to see your own choices. Use the identifiers it supplies if you set --model; display names are not a reason to invent a command-line model ID.

Install on Windows, macOS, or Linux

Google's hands-on codelab, read October 26, gives these official one-liners.

macOS and Linux, in a terminal:

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

Windows, in PowerShell:

irm https://antigravity.google/cli/install.ps1 | iex

Those commands execute the downloaded installer immediately. For a business workstation, we prefer a visible pause between downloading a script and running it. The following version saves the same official file so you or your tech person can read it first.

Download, read, then run on macOS or Linux

Run this first block:

agy_installer=$(mktemp)
curl -fsSL https://antigravity.google/cli/install.sh -o "$agy_installer"
less "$agy_installer"

Press q to leave the reader. If the download succeeded and you are satisfied with the script, run the next block in the same terminal:

bash "$agy_installer"
rm -f "$agy_installer"

Run this as the person who will use agy. Its normal destination is that person's home directory, so installing it under a different login can leave your working terminal unable to find it.

Download, read, then run on Windows

In PowerShell, download a uniquely named temporary script and open it:

$agyInstaller = Join-Path $env:TEMP ("agy-install-" + [guid]::NewGuid() + ".ps1")
Invoke-WebRequest -Uri https://antigravity.google/cli/install.ps1 -OutFile $agyInstaller
notepad $agyInstaller

After reading the file, run this in the same PowerShell window if you want to proceed:

powershell -NoProfile -ExecutionPolicy Bypass -File $agyInstaller
Remove-Item -LiteralPath $agyInstaller

This sets the execution policy for that PowerShell process. It does not ask you to permanently change the machine's policy. The filename also makes it clear which downloaded script you are running.

What the installer changes

We read both official installers on October 26. They identify the platform, request a release manifest, download the named release, and compare its SHA-512 checksum with the manifest. A mismatch stops installation with a security error. That check tests whether the downloaded payload matches the release manifest; reading the script remains a separate review step.

The usual destinations are:

Scroll to compare columns and rows when the table extends beyond the box.

System Binary location
macOS and Linux $HOME/.local/bin/agy
Windows %LOCALAPPDATA%\agy\bin\agy.exe

The scripts then run the binary's install command to configure the shell. In our tests, Windows added its bin directory to the user's PATH. The Unix setup added a PATH entry to .bashrc and .profile. Shell setups differ, so read any PATH warning at the end, especially on a Mac using zsh.

Open a new terminal, then check:

agy --version

If the command is not found, check the destination above before downloading again. On Unix you can run "$HOME/.local/bin/agy" --version. In PowerShell you can run & "$env:LOCALAPPDATA\agy\bin\agy.exe" --version. Our kickstart scripts use those locations to check the installation before a new PATH has taken effect.

Sign in once, with the browser ready

The tested login flow gives you 60 seconds. Have a browser ready before starting. Run the login in a real terminal, even if the later workload will be automated.

  1. Open the terminal where you intend to use the CLI and run agy.
  2. Open the Google login URL it prints. Choose the intended account and check the application name before approving sign-in.
  3. The browser returns an authorization code. Use the Copy to Clipboard button.
  4. Return to the terminal. Paste the code at its authorization-code prompt and press Enter before the 60-second wait expires.
  5. If the attempt expires, start again and use the new link and code. After sign-in, try a small prompt.

Our smoke test was:

agy -p "Reply with the single word OK"

It returned OK on both tested systems. That checks the login and a request, without asking the agent to modify a project.

For SSH or a machine without a browser, open the printed URL on another device. Complete sign-in there, then paste the code back into the SSH terminal. We verified that flow. Piping a code into print mode did not work in our test; use the terminal prompt for this first step.

In the tested release, the saved login was under the user's home directory at .gemini/antigravity-cli/antigravity-oauth-token. On Linux its permissions were mode 600. Treat it like a password: do not put it in a repository, support ticket, shared archive, or screenshot. Login storage may change between releases; our observed location is a reason to protect that directory, not to assume it is the only possible credential store.

Remove the old Gemini CLI carefully

If you used the npm-installed Gemini CLI only with a personal Google login, remove that package with:

npm uninstall -g @google/gemini-cli

Use the same account and npm environment that installed it. If you keep Gemini CLI for a paid API key or an existing business license, skip this cleanup. You can assess Antigravity separately.

After retiring the old personal-login setup, its saved login files are .gemini/oauth_creds.json and .gemini/google_accounts.json under your home directory. On macOS or Linux, remove just those files:

rm -f "$HOME/.gemini/oauth_creds.json" "$HOME/.gemini/google_accounts.json"

In PowerShell:

Remove-Item -LiteralPath "$HOME\.gemini\oauth_creds.json", "$HOME\.gemini\google_accounts.json" -ErrorAction SilentlyContinue

Do not delete the whole .gemini folder. Antigravity CLI keeps data beneath it too, and you may have instructions or configuration worth retaining. Removing a program and removing all nearby folders are different operations. Our kickstart scripts offer the npm uninstall; they do not delete these login files for you.

What changed in our kickstart scripts

The Windows download, macOS download, and Linux download now offer Google Antigravity CLI behind a default-no prompt. Unattended --auto runs, or -AutoMode on Windows, keep skipping this optional assistant.

The scripts first look for an existing agy and report its version. If you choose installation, they download Google's installer to a temporary file and run that file, with output sent to the existing kickstart log. Their health checks and command summaries now use agy.

They also check for the old global npm package. If it is present, they explain the personal-account cutoff and offer removal, default yes in an interactive session. Automatic runs only print the notice. They leave the installed package in place so an unattended update does not remove someone's working API or business-license setup.

The kickstart article contains the same scripts as the downloads. Its dated note links here, and the earlier posts now tell personal-account readers where the install instructions changed.

First useful commands

Once signed in, these are small checks you can run before assigning real work:

agy models
agy -p "Draft a five-item website review checklist for a local shop." --output-format json
agy mcp list
agy changelog
agy update

The JSON example requests a draft and makes no claim about your actual site. Check the returned structure before connecting it to another program. JSON output helps software consume a response; it does not verify the response's facts.

For a first project session, start in a copy of a small work folder and ask for a plan. Describe the intended result, what files are in scope, and how you will check the answer. Our after-install workflow develops that habit. The $97 Launch covers the broader small-business practice of building something modest, checking it, and improving it from real use.

FAQ

Does paying for Google AI Pro keep Gemini CLI working?

Not through the affected personal-account login route. Google includes Pro and Ultra in the June 18 cutoff. A paid Gemini API key is a different access route.

Should I reinstall Gemini CLI or try another personal account?

That will not reverse the account-tier cutoff. Move the personal-account workflow to Antigravity CLI. If your error is different, diagnose that error rather than assuming every failure has this cause.

Do I need Node.js for Antigravity CLI?

Not for the native CLI installation described here. Other tools and MCP servers in your workflow may still need Node.js, so do not remove it just because this CLI no longer uses npm.

Can I sign in over SSH?

Yes. Open the printed login URL in a browser on another device and paste the returned code into the SSH terminal within the tested 60-second window.

Can I keep both CLIs installed?

Yes. Keeping Gemini CLI can make sense for a paid API key or an existing Code Assist business license. Check the separate business subscription timeline before relying on that license long term.

Will I get the same models and features?

Run agy models to check your account. Our account showed Gemini, Claude, and GPT-OSS models, but plans may differ. Google did not promise one-to-one feature parity at launch, so test your own workflow.

Fact-check notes and sources

All web sources below were read on October 26, 2026. Installation, login, command-help, and model observations came from our Windows 11 workstation and Linux desktop container tests that day. We tested version 1.4.7; those observations are not a guarantee for other plans or later versions.

Related reading

← Back to Blog

Accessibility Options

Text Size
High Contrast
Reduce Motion
Reading Guide
Link Highlighting
Accessibility Statement

J.A. Watte is committed to ensuring digital accessibility for people with disabilities. This site conforms to WCAG 2.1 and 2.2 Level AA guidelines.

Measures Taken

  • Semantic HTML with proper heading hierarchy
  • ARIA labels and roles for interactive components
  • Color contrast ratios meeting WCAG AA (4.5:1)
  • Full keyboard navigation support
  • Skip navigation link
  • Visible focus indicators (3:1 contrast)
  • 44px minimum touch/click targets
  • Dark/light theme with system preference detection
  • Responsive design for all devices
  • Reduced motion support (CSS + toggle)
  • Text size customization (14px–20px)
  • Print stylesheet

Feedback

Contact: jwatte.com/contact

Full Accessibility Statement • Privacy Policy

Last updated: April 2026