Gemini CLI in 2026: Who Can Still Use It, and How to Migrate to Antigravity CLI (agy)

Updated 2026-10-26: We rewrote this guide around the June 18, 2026 cutoff. It now covers who can still use Gemini CLI, how to migrate to Antigravity CLI (agy), and what the Gemini 4 Argon launch and the September 2026 limits on Gemini 2.5 access mean for you. The old free-tier and "Gemini 2.5 Pro by default" claims have been removed.

Earlier versions of this guide called Gemini CLI the cheapest way to try a terminal coding agent: install with npm, sign in with a personal Google account, and get a generous free daily quota. That is no longer true. On May 19, 2026, Google announced the move from Gemini CLI to Antigravity CLI. Consumer access ended on June 18, 2026.

If gemini stopped answering on your personal account, nothing is wrong with your install. Your account type is no longer served. If you're on an enterprise Code Assist license or bill through a paid API key, Gemini CLI still works and is still maintained. The npm package @google/gemini-cli shipped stable version 0.65.0 on October 20, 2026, and nightly builds are still being published.

This guide covers:

• The current status of Gemini CLI.
• Which account types still work.
• How to install and authenticate if you're one of them.
• What API-key usage costs.
• A practical migration path to Antigravity CLI for everyone else.

Gemini CLI status: what changed and when

What Gemini CLI does

For teams that still have access, the tool itself hasn't changed. Per the README and the Gemini CLI docs, Gemini CLI is an open-source terminal agent (Apache License 2.0) that can:

• Read, write and edit files across a codebase.
• Run shell commands, asking you to approve each tool call unless you turn on auto-approve.
• Search the web via Google Search (the google_web_search tool returns a summary with source URLs) and fetch web pages.
• Connect to MCP servers for custom tools.
• Load project instructions from GEMINI.md files.

User settings live in ~/.gemini/settings.json. Project overrides go in .gemini/settings.json at the repo root. There is also a system-wide file, for example /etc/gemini-cli/settings.json on Linux, which overrides both.

Who can still use Gemini CLI

Whether Gemini CLI works for you depends only on your account type. Here is the breakdown from Google's announcement and the Gemini CLI docs, as of October 2026:

Google's commitment covers paid Gemini API keys. The announcement doesn't promise anything for free-tier API keys, so don't build a workflow that depends on Gemini CLI plus a free key.

Gemini CLI not working? Check these first

• Personal account, requests fail since mid-June. This is the June 18 cutoff, not a bug. Reinstalling or clearing ~/.gemini won't help. Install Antigravity CLI (steps below) or switch Gemini CLI to a paid API key.
• Company or Workspace account, sign-in works but requests fail. For organization accounts and some Code Assist licenses, the Gemini CLI docs require a Google Cloud project. Run export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID", make sure the Gemini for Cloud API is enabled and your IAM permissions are set, then pick Sign in with Google again.
• A Gemini 2.5 model returns access errors on a new project. Since September 18, 2026, Google only serves the 2.5 models to users who have used them before. Choose a Gemini 3.x model instead.
• You want Gemini 4 Argon. The README says Gemini CLI "has not been optimized for Gemini 4 Argon" and tells you to use Antigravity for it.
• Vertex AI auth picks the wrong credentials. The docs tell you to run unset GOOGLE_API_KEY GEMINI_API_KEY before using ADC or a service account, so a stray key doesn't take priority.
• command not found: gemini. npm's global bin directory is not on your PATH. Run npm config get prefix and add $prefix/bin to your PATH.

Which models Gemini CLI uses now

Earlier versions of this post said Gemini 2.5 Pro was Gemini CLI's default model. That is out of date. The current Gemini CLI model docs don't name a single default. The /model dialog offers two routing modes, Auto (Gemini 3) and Auto (Gemini 2.5), plus a Manual picker, and Google recommends one of the Auto options. Three things limit what you can actually pick:

• Gemini 2.5 is legacy-only on the API. Since September 18, 2026, only users who used the 2.5 models before can access gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite. Google says they are not deprecated and have no announced shutdown date.
• Gemini 3.8 Flash is Google's current recommended Flash model. It went GA on September 2, 2026. Since October 8, requests to gemini-3.7-flash are automatically routed to it. See our Gemini 3.8 Flash guide for details.
• Gemini CLI isn't optimized for Gemini 4 Argon. Google's flagship went GA on October 13, 2026 on the paid API and in Antigravity, and the Gemini CLI README directs Argon users to Antigravity. Specs and pricing are in our Gemini 4 Argon guide.

Which models you see in /model depends on your account and auth method. To pin a model for one run, use -m (the README's example is gemini -m gemini-2.5-flash), and confirm the model is available to your account first.

How to install Gemini CLI (supported accounts)

Install Gemini CLI only if you're on an enterprise Code Assist license or using a paid API key or Vertex AI. If you have a personal account, skip to the Antigravity CLI migration.

Prerequisites

• Node.js 20 or newer. The current npm package requires node >=20. Check with node --version.
• One supported credential: a Code Assist Standard or Enterprise seat, a paid Gemini API key, or a Google Cloud project with Vertex AI enabled.

Install on macOS, Linux or Windows

npm install -g @google/gemini-cli
gemini --version

The README also lists Homebrew, MacPorts and Anaconda installs. If you just want to try it once, run it without installing:

npx @google/gemini-cli

If npm install -g fails with EACCES, install Node through nvm, or run npm config set prefix ~/.npm-global and add ~/.npm-global/bin to your PATH.

Gemini CLI authentication in 2026

1. Sign in with Google (organization accounts and Code Assist licenses)

export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
gemini   # choose "Sign in with Google"

The docs say a project is required for company, school and Workspace accounts and for some Gemini Code Assist licenses. Gemini CLI reads GOOGLE_CLOUD_PROJECT first and falls back to GOOGLE_CLOUD_PROJECT_ID. In PowerShell, use $env:GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID".

2. Gemini API key (usage-based billing)

export GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
gemini   # choose "Use Gemini API key"

Create the key in Google AI Studio. No Google Cloud project is required. For long-term support, this should be a key on a paid billing account. Put the export in your shell profile or a secrets manager rather than typing it on the command line, so it doesn't end up in your shell history.

3. Vertex AI (Gemini Enterprise Agent Platform)

All Vertex methods need a project and a location:

export GOOGLE_CLOUD_PROJECT="YOUR_PROJECT_ID"
export GOOGLE_CLOUD_LOCATION="YOUR_PROJECT_LOCATION"

# Option A: Application Default Credentials
unset GOOGLE_API_KEY GEMINI_API_KEY
gcloud auth application-default login

# Option B: service account with the "Vertex AI User" role
unset GOOGLE_API_KEY GEMINI_API_KEY
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/keyfile.json"

# Then run gemini and choose "Vertex AI"

There is also a third option, a Google Cloud API key set as GOOGLE_API_KEY. If you get errors saying API keys aren't supported, your organization's policy probably blocks them, so use ADC or a service account instead. Vertex is the right path when you need org-managed quota, IAM and audit logs.

Day-to-day usage

Run gemini inside a project directory and describe the task in plain English. For example: "Add a Postgres health-check endpoint to server.ts and a test in tests/health.spec.ts." The agent reads files, proposes edits and shell commands, and asks you to approve each tool call.

• gemini -m <model-id> pins a model for the session.
• gemini --yolo auto-approves tool calls. Only use it inside a disposable container or VM.
• /model switches between the Auto and Manual model options.
• /memory views or edits persistent project memory, which comes from GEMINI.md.
• /auth reruns authentication, for example to move from a personal login to an API key.
• /help lists the commands available in your installed version.

MCP servers in Gemini CLI

Gemini CLI keeps MCP servers inline in ~/.gemini/settings.json (global) or .gemini/settings.json (per project):

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/code"]
    }
  }
}

Antigravity CLI moves MCP config out of settings.json into a separate file. That change is covered in the migration section below. For server ideas, see our roundup of MCP servers for coding agents.

What Gemini CLI costs on a paid API key

With an API key, Gemini CLI has no subscription fee. You pay Gemini API token rates, and agent sessions use more tokens than you might expect because the agent re-sends file context with each step. Here is a worked example for one heavy session using 2,000,000 input tokens and 200,000 output tokens, at standard paid-tier rates from Google's pricing page as of October 2026, with no caching:

Output prices include thinking tokens. Gemini 2.5 Pro charges higher rates ($2.50 / $15.00) on prompts over 200K tokens. Context caching cuts the input side a lot: cached input on gemini-3.8-flash is $0.075 per 1M through December 31. Take the 3.8 Flash price increase on January 1, 2027 into account before you set a team budget.

How to migrate from Gemini CLI to Antigravity CLI (agy)

Antigravity CLI is Google's replacement terminal agent. According to Google, it is built in Go and shares its agent harness with the Antigravity 2.0 desktop app. It is designed for background, asynchronous multi-agent work. Google's migration guide names the launcher command agy. Third-party guides, including InventiveHQ and Avinash Sangle, report that it is closed source, unlike Gemini CLI's Apache-2.0 codebase. Google's own pages don't say either way. Google has also said up front that "there won't be 1:1 feature parity right out of the gate."

Step 1: Install Antigravity CLI

These commands come from Google's download page. Antigravity CLI is a new binary, not an npm update of Gemini CLI.

# macOS, Linux, Googlebook (Linux terminal)
curl -fsSL https://antigravity.google/cli/install.sh | bash

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

Run agy and sign in with the same Google account you used for Gemini CLI. InventiveHQ reports that the macOS/Linux installer puts the binary at ~/.local/bin/agy, so add that directory to your PATH if agy isn't found.

Step 2: Let the first-launch checklist convert your config

Google's migration guide says that when you run agy on a machine with Gemini CLI config, it shows an interactive checklist. The checklist offers to convert extensions and global config, move session tokens to the OS keyring, and map visual settings. Parity is partial, so some Gemini CLI themes and visual overlays won't come across.

Step 3: Convert extensions to plugins

Gemini CLI extensions become Antigravity plugins. To run the conversion explicitly:

agy plugin import gemini

Google says the command searches your legacy local directories, parses extension manifests and converts the files. In Google's sample output, legacy custom commands become skills and MCP server definitions move into mcp_config.json. Avinash Sangle reports two gaps. Extensions that rely on Node.js-only APIs trigger a warning during import, and custom themes bundled inside extensions are dropped silently. InventiveHQ adds that such Node.js-dependent extensions need a Go rewrite or the official MCP bridge.

Step 4: Keep GEMINI.md and AGENTS.md as they are

You don't have to rewrite your rules files. Google's guide says both CLIs use the same workspace context rules. Antigravity CLI reads GEMINI.md and AGENTS.md from the active directory and global rules from ~/.gemini/GEMINI.md. If your team keeps one AGENTS.md for several agents, it keeps working.

Step 5: Move skills, MCP config and hooks to the new paths

The biggest change is folder layout. Workspace-level files move from .gemini/ to .agents/, and global CLI files move under ~/.gemini/antigravity-cli/ or ~/.gemini/config/. These paths come from Google's migration and settings docs:

Workspace skills aren't picked up until they're in .agents/skills/. In a repo, use git mv so the history is kept:

mkdir -p .agents
git mv .gemini/skills .agents/skills

If you use Gemini CLI's top-level agents settings block (for example agents.overrides), check it by hand after conversion. Antigravity CLI's settings docs don't list an agents key. The closest setting is agentMode, which controls the execution mode for file edits, commands and planning. None of the guides we checked documents a one-to-one mapping.

Step 6: Rename url to serverUrl for remote MCP servers

This change is easy to miss. Google's guide says remote (SSE/WebSocket) MCP servers that used url or httpUrl in Gemini CLI must use serverUrl in Antigravity CLI. A Gemini CLI entry like this:

"internal-docs": {
  "url": "https://mcp.internal.example.com/sse"
}

becomes this in mcp_config.json:

"internal-docs": {
  "serverUrl": "https://mcp.internal.example.com/sse"
}

InventiveHQ reports that a leftover url key doesn't cause an error at startup. The failure only shows up when the agent first calls a tool from that server, so test each remote server after migrating. Keep whatever top-level structure the importer writes to mcp_config.json, and check it against Google's docs for your installed version.

Step 7: Update scripts and CI

• Pipelines and cron jobs that call gemini are not migrated automatically. Find them with grep -rn "gemini " .github/ scripts/ and decide, job by job, whether each one moves to agy or stays on Gemini CLI with a paid API key.
• For headless CI, Gemini CLI with a paid API key or Vertex credentials is still the best-documented option. The Antigravity settings docs list a modelProvider setting that accepts "gemini" and reads GEMINI_API_KEY, but check API-key auth on your version before you rely on it in a pipeline.
• InventiveHQ reports that Gemini CLI's ACP stdio mode (gemini --acp) was reportedly absent from Antigravity CLI at launch and is tracked in a GitHub issue. Per-call temperature, top_k and system-instruction overrides are also gone, per Avinash Sangle.

A note on quotas

Google's announcement doesn't publish Antigravity CLI quotas. InventiveHQ reports that the old ~1,000-requests-per-day quota was replaced by a weekly compute-based cap, and that parallel subagents each use their own token budget. Avinash Sangle says the weekly cap is the biggest functional difference between Pro and Ultra. Check Antigravity's pricing page for your plan before you move a team's daily workflow over.

Gemini CLI vs Antigravity CLI at a glance

Should enterprise teams stay on Gemini CLI or move now?

If you're on Code Assist Standard or Enterprise, nothing forces you to move today. Google says it will keep supporting enterprise customers on Gemini CLI "with access to the latest Gemini models and other updates." It doesn't say which models, though, and Gemini CLI isn't optimized for Gemini 4 Argon. Here is a practical way to decide:

• Stay on Gemini CLI for now if you rely on an open-source, auditable agent, if you have CI built on gemini with API-key or Vertex auth, or if you use features like ACP that InventiveHQ reports were absent from Antigravity CLI at launch.
• Move to Antigravity CLI if your developers want Gemini 4 Argon in the terminal, run parallel background agents, or are on personal Pro or Ultra plans that Gemini CLI no longer serves.
• Either way, start with one repo. Migrate a single service, run agy plugin import gemini, move .agents/ files, fix every serverUrl, and run your usual tasks before you roll it out further.

If you're deciding between Google's tooling and other vendors' agents, our Gemini CLI vs Claude Code comparison and Cursor vs Claude Code vs Codex vs Gemini CLI breakdown cover the alternatives. Our Claude Code vs OpenAI Codex guide covers the two biggest non-Google terminal agents.

Security notes for both CLIs

• Auto-approve modes, such as Gemini CLI's --yolo or Antigravity's auto agent mode, will run whatever shell commands the model produces. Use them only in disposable sandboxes.
• Treat ~/.gemini/ like an SSH key directory. It holds auth state and, after migration, Antigravity CLI config too. Google's guide says Antigravity moves session tokens to the OS keyring.
• Remote MCP servers often carry tokens in env blocks inside mcp_config.json. Keep that file out of git, or reference secrets from your environment.
• Review converted plugins before you trust them. The import rewrites commands as skills, and you should read what it generated.

FAQ

Is Gemini CLI deprecated?

It is retired for individuals but not discontinued. Since June 18, 2026, it no longer serves Google AI Pro, Ultra or free Code Assist for individuals users. It is still supported and updated for Gemini Code Assist Standard and Enterprise licenses and for paid Gemini API and Agent Platform keys. Enterprise Code Assist for GitHub, the separate PR review app, is also unaffected.

Is Gemini CLI still free?

Not for personal accounts. The free individual tier was one of the account types cut off on June 18, 2026. You can still use Gemini CLI with a paid API key, billed per token, or through an enterprise Code Assist license.

Why did Gemini CLI stop working on my account?

If you sign in with a personal Google account or a Google AI Pro or Ultra subscription, Google stopped serving your requests on June 18, 2026. Install Antigravity CLI (agy), or switch Gemini CLI to a paid Gemini API key with /auth.

What is agy?

agy is the command for Antigravity CLI, Google's replacement for Gemini CLI. Google says it is built in Go, supports background multi-agent workflows, and shares its harness with the Antigravity 2.0 desktop app. Third-party guides report that it is closed source.

Do GEMINI.md and AGENTS.md work in Antigravity CLI?

Yes. Google's migration guide says both CLIs use the same workspace context rules. Antigravity CLI reads GEMINI.md and AGENTS.md in the active directory, plus ~/.gemini/GEMINI.md for global rules.

What happens to my Gemini CLI extensions?

They become Antigravity plugins. Run agy plugin import gemini, or accept the conversion on first launch. Legacy commands are converted into skills and MCP servers move into mcp_config.json. Third-party guides (Avinash Sangle, InventiveHQ) say extensions that use Node.js-only APIs may need rework.

Can Gemini CLI use Gemini 4 Argon?

Not officially. The README says Gemini CLI "has not been optimized for Gemini 4 Argon" and tells you to use Antigravity. Argon is available in the Antigravity desktop app and CLI for Google AI Pro and Ultra subscribers and eligible Gemini Enterprise subscriptions, and on the paid Gemini API.

Is Gemini 2.5 Pro still Gemini CLI's default model?

No. The current docs offer Auto (Gemini 3) and Auto (Gemini 2.5) routing plus manual selection, and don't name a single default. Since September 18, 2026, the Gemini API only gives Gemini 2.5 access to users who have used those models before.

Why did my remote MCP server stop working after migrating?

Most likely the key name. Antigravity CLI expects serverUrl where Gemini CLI used url or httpUrl. InventiveHQ reports that the mismatch only shows up when the agent calls a tool from that server, not at startup.

Sources

• Google Developers Blog: Transitioning Gemini CLI to Antigravity CLI (May 19, 2026)
• Google Antigravity docs: Migrating from Gemini CLI
• Google Antigravity docs: CLI settings
• Google Antigravity: Download and install
• Google Antigravity blog: Gemini 4 Argon in Google Antigravity
• Gemini CLI README (GitHub)
• Gemini CLI docs: Authentication
• Gemini CLI docs: Model selection
• Gemini CLI docs: Configuration
• Gemini CLI docs: Web search tool
• npm: @google/gemini-cli
• Gemini API: Deprecations
• Gemini API: Changelog
• Gemini API: Pricing
• InventiveHQ: Gemini CLI deprecated, Antigravity CLI migration
• InventiveHQ: Antigravity CLI install and config migration
• Avinash Sangle: Gemini CLI to Antigravity CLI migration guide
• AI Builder Club: Antigravity CLI guide

If your team is planning a migration from Gemini CLI or standardizing on terminal agents, Codersera can help you hire vetted remote developers who have shipped agent-assisted work and can own the review and rollout.