Is Gemini CLI Deprecated? Shutdown Date and What to Use Now

Google shut down Gemini CLI for free, Pro, and Ultra users on June 18, 2026. Any CI/CD pipeline, shell script, or automation still calling the gemini command is already failing, with no grace period behind it. The replacement is Antigravity CLI - a closed-source Go rewrite that did not have feature parity at launch. For individual builders, the practical migration path is Claude Code for terminal workflows or Aider for full open-source control.

Is Gemini CLI Deprecated?

Yes. Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests on June 18, 2026 for Google AI Pro, Ultra, and anyone using Gemini Code Assist for individuals free of charge. That date has passed. This is finished, not upcoming.

Two groups were never cut off, which is the part that confuses everyone:

• Gemini Code Assist Standard or Enterprise licenses. Google's announcement says access "remains unchanged" for organizations on those licenses.
• Paid API keys. Gemini CLI "will remain accessible via paid Gemini and Gemini Enterprise Agent Platform API keys."

So when a colleague tells you Gemini CLI still works for them, they are probably right. It works for them.

If you are hitting a sign-in error that starts with "This client is no longer supported for Gemini Code Assist for individuals" and points you at antigravity.google, that is the shutdown, not a broken install. Reinstalling, wiping your config, or signing in with a different personal Google account will not bring it back, because every individual tier lost access on the same day. Your three real options are an enterprise license, a paid API key, or a different CLI. The rest of this page covers the third one.

What Did Google Announce About Gemini CLI?

On May 19, at Google I/O 2026, Google announced that Gemini CLI and Gemini Code Assist IDE extensions would stop serving requests on June 18 for Google AI Pro, Ultra, and free-tier users. Individual Gemini Code Assist subscribers were cut off too. Gemini Code Assist for GitHub stopped taking new org installations on the same date, and the consumer version has since been sunset, with code review activity ceasing.

Enterprise Gemini Code Assist Standard and Enterprise license holders keep access indefinitely. Gemini CLI also remains accessible via paid Gemini and Gemini Enterprise Agent Platform API keys.

The replacement is Antigravity CLI - Google's "premier agent-first development platform." The transition is mandatory for non-enterprise users. There is no opt-out.

Why Did Google Accept 6,000 Contributions Then Lock the Tool?

This is the part that made developers angry. Before the shutdown announcement, Google accepted over 6,000 merged pull requests from external contributors to the Apache 2.0 Gemini CLI repository over nearly a year. The project accumulated 100,000+ GitHub stars. Google's own announcement cited those contributions as evidence of the project's success - then used that success to justify consolidating into a closed-source replacement.

Developer Andrea Alberti had a 27-commit pull request merged the same day as the shutdown announcement. Her response: "Essentially working for free on a code base that will only be used in enterprises?"

The Hacker News thread captures the pattern clearly. Developers do not object to Google building a better enterprise product. They object to community investment in an open-source project being used to improve a product that then excludes the contributors. Enterprise customers keep Gemini CLI plus Antigravity CLI access. Everyone who built the tool gets neither.

The Linux Foundation spotlighted this exact dynamic at Open Source Summit North America 2026, introducing a Model Openness Tool designed to identify this kind of pattern. The Gemini CLI shutdown became exhibit A.

What Is Antigravity CLI and How Does It Compare?

Antigravity CLI is Google's proprietary replacement, invoked with the `agy` command. It is a full rewrite - not a port - of the Gemini CLI codebase.

Feature	Gemini CLI	Antigravity CLI
Language	TypeScript/Node.js	Go (single binary)
License	Apache 2.0 (open source)	Proprietary (closed source)
Individual user access	Ended June 18, 2026	Available now
Multi-agent support	Limited	Native async multi-agents
Feature parity	Full (mature)	Incomplete at launch
Extensions	Plugin system	Hooks + plugin import from Gemini
Context file	GEMINI.md	AGENTS.md (also still reads GEMINI.md)

Google's own announcement states: "There won't be 1:1 feature parity right out" at launch. Subagents and extensions are being brought over, but other features are still in progress. For individual builders who relied on Gemini CLI for terminal-based coding workflows, Antigravity CLI is a step backward in the short term.

The install is straightforward (`curl -fsSL https://antigravity.google/cli/install.sh | bash`) and you can import existing Gemini plugins with `agy plugin import gemini`. Both binaries coexist during the migration window.

What Should You Do Now That Gemini CLI Is Off?

The deadline has already passed, so there is no runway left to plan around. Priority order if you are still on it:

1. Audit your automation immediately. Search CI/CD pipelines, GitHub Actions, shell scripts, and cron jobs for any reference to the `gemini` command. Every one of these is already failing with no graceful degradation - just errors.

2. Categorize your usage. Most Gemini CLI usage falls into three buckets:

• Interactive terminal coding sessions - the most common use case
• Scripted code generation - via `gemini -p "prompt"` in automation
• IDE extensions - Gemini Code Assist in VS Code/JetBrains

Each bucket has a different optimal migration path.

3. Migrate by category. For interactive terminal work, Claude Code is the closest analog. For scripted tasks, `claude -p` replaces `gemini -p`. For IDE integration, the Claude Code VS Code extension replaces Gemini Code Assist. If you want open-source provider flexibility, Aider handles all three.

4. Test one real workflow before you trust it. Not a toy example. Run your actual daily workflow through the new tool. This is the step most builders skip, and it surfaces 90% of the migration friction.

Which Alternative Should You Choose?

Most migration guides hedge with a long list. Here is a direct recommendation based on use case:

For terminal-based coding sessions: Claude Code. Same terminal-first workflow as Gemini CLI - file editing, bash execution, multi-file context, and now parallel subagents via Dynamic Workflows. The beginner setup takes about 20 minutes. If you used `gemini -p` for scripted tasks, the equivalent is `claude -p`.

For IDE integration: The Claude Code VS Code extension replaces Gemini Code Assist's IDE functionality. Alternatively, Cursor provides a full AI-native IDE experience.

For full open-source control: Aider works with any LLM provider (Anthropic, OpenAI, Google, local models), has strong Git integration, and has been stable longer than most alternatives. This is the right choice if you do not want to depend on any single AI vendor.

Other options worth knowing about: OpenCode (161K+ GitHub stars, open source, just shipped Desktop v2 with push-based background agents) and Continue (VS Code extension, multiple providers). But Claude Code has the most feature overlap with Gemini CLI and the largest active community.

What Does the Gemini CLI Shutdown Tell You About Tool Risk?

The incentive structure behind this shutdown is worth understanding before you pick your next tool. The risk factors that led here:

• Open-source wrapper around a closed commercial core. The code was Apache 2.0, but it depended on Google's cloud backend. Google controlled the on/off switch the entire time.
• Enterprise revenue always wins. When the enterprise tier generates the majority of revenue, individual users are expendable. Google made this explicit by letting enterprise customers keep both tools.
• No contractual commitment to free access. Terms of service for free products can change with a blog post. They did.

Claude Code has different characteristics. It is Anthropic's flagship developer product and a core revenue driver - Anthropic raised $65B in Series H funding at a $965B valuation in May 2026, with run-rate revenue crossing $47B. Claude Code is central to that revenue story.

The June 15 billing changes (separate Agent SDK credits for `claude -p` and programmatic usage) show Anthropic expanding the business model - splitting interactive and programmatic pools to serve both audiences better - not restricting access. Interactive Claude Code terminal usage remains unchanged on your subscription.

That is not a guarantee of anything. Google's developer tools have followed the deprecation arc before: Firebase free tier changes, Cloud Run, Google Code, Stadia, Google+ API. The point is not that one vendor is trustworthy and another is not. The point is: choose tools where your use case aligns with the vendor's revenue model. Right now, for individual AI builders doing terminal-based coding, Anthropic's incentive structure is more aligned than Google's.

The Broader Pattern for AI Builders

Gemini CLI is not the first tool to follow this arc, and it will not be the last. Three principles for choosing tools in this environment:

1. Check where revenue comes from. If individual users are the product's revenue center, you are the customer. If enterprise is the revenue center, you are the funnel. Funnels get optimized away.

2. Distinguish open-source code from open-source access. Gemini CLI's code is still Apache 2.0. But the code without the API backend is useless. Real openness means the tool works without a single vendor's permission.

3. Maintain tool optionality. Never have exactly one tool in a critical workflow path. The builders who get hurt by shutdowns like this are the ones who went all-in on a single vendor with no fallback. Keep your workflows modular enough to swap providers in a week, not a quarter.

If you are still working through the migration, the Claude Code beginner guide covers the full setup. The transition is straightforward for most terminal-based workflows - the command patterns, file editing model, and context handling are similar enough that muscle memory transfers within a few sessions.