The short answer
Gemini CLI is Google’s coding agent for the terminal. You give it a task in plain language, and Gemini CLI reads your project’s code, edits files, runs commands and looks things up on the web when it needs to. Google’s Gemini models power the agent.
Gemini CLI is open source, under the Apache 2.0 license. You install it with npm or Homebrew and run it in your project folder.
There is one thing to know before you start. Since 18 June 2026, Google sends anyone who works with a personal Google account, free or with Google AI Pro or Ultra, to a successor: Antigravity CLI. Gemini CLI itself remains for companies with Gemini Code Assist, and for anyone who works with an API key or through Vertex AI in Google Cloud.
What Gemini CLI does
A task like “add a search field to the customer list” gets broken into steps. Gemini CLI finds the files involved, writes the change and runs the tests. You watch and approve actions before they happen.
Working with files and the shell
Gemini CLI reads and writes files in your project and runs commands in your terminal, such as tests or a build. Any action that changes something asks for permission first.
Searching the web
The agent can fetch a web page or look something up with Google Search, such as a library's documentation. That way it does not work only from what the model already knows.
Conventions in GEMINI.md
A GEMINI.md file holds your project's conventions: how you test, which style you follow, what is off limits. Gemini CLI reads that file in every session.
MCP and extensions
MCP servers connect extra tools, such as your issue tracker or your database. Extensions bundle such tools with their own commands, so you install them in one go.
Without a screen, in scripts
The -p option passes a task without a chat window, and --output-format json returns the answer as JSON. That is how you put Gemini CLI in a script or a CI pipeline.
Gemini CLI can also run commands in a shielded environment, a sandbox, so a mistake does not touch your computer. That combination of model, tools and limits is called an AI harness, explained in what is an AI harness.
Who still uses Gemini CLI, and who uses Antigravity CLI?
This is the question most explanations online do not answer yet. Google has split the way in.
Gemini CLI is for:
- teams with Gemini Code Assist Standard or Enterprise, or Gemini Code Assist for GitHub;
- anyone who works with a Gemini API key from Google AI Studio;
- anyone who uses the models through Vertex AI in Google Cloud.
Antigravity CLI is for anyone who signs in with a personal Google account, on the free tier or with Google AI Pro or Ultra. The command is agy. Antigravity CLI shares its agent with Google’s Antigravity desktop app, so one sign-in covers the terminal, the app and your editor. A Workspace account from work does not give access to Antigravity CLI.
Switching takes little work. On first launch, Antigravity CLI offers to take over your Gemini CLI settings. Your GEMINI.md and AGENTS.md work unchanged, and the command agy plugin import gemini turns your extensions into plugins. MCP servers then live in a file of their own, mcp_config.json. According to Google, Antigravity CLI cannot yet do everything Gemini CLI does, so test your own workflow before you switch.
What does Gemini CLI cost?
Gemini CLI itself is free. What you pay depends on how you get access.
A free API key from Google AI Studio lets you try it, but only with a Flash model and a daily limit. Since March 2026 the Pro models are paid only. With a paid API key or through Vertex AI you pay per use, and with Gemini Code Assist it comes with your team’s license.
Antigravity CLI runs on your Google account’s subscription. The free tier gives an amount of use that refills every week, with a Flash model. Google AI Pro and Ultra give more use and access to the stronger models. Google adjusts the limits and models regularly, so check Google’s own pages for current figures.
Gemini CLI, Claude Code and Codex side by side
All three are coding agents you hand a task. The difference lies mostly in who supplies the models and what you already pay for.
| Gemini CLI | Claude Code | Codex | |
|---|---|---|---|
| Made by | Google. | Anthropic, the company behind Claude. | OpenAI, the company behind ChatGPT. |
| Models | Google's Gemini models. | Anthropic's Claude models. | OpenAI's models, locally also an open model. |
| Where you work | Terminal. For individuals now through Antigravity CLI, with a desktop app and editor alongside. | Terminal, VS Code, JetBrains, desktop app or browser. | Terminal, editor, ChatGPT in the browser or the desktop app. |
| Conventions in | GEMINI.md | CLAUDE.md | AGENTS.md |
| Open source | Yes, under Apache 2.0. | No. | The CLI is, the cloud version is not. |
| Cost | Through Gemini Code Assist, an API key or Vertex AI. For individuals through your Google subscription in Antigravity CLI. | Included in the paid Claude plans, or per use through the API. | Limited on the free plan, full from ChatGPT Plus, or per use through the API. |
Choose Gemini CLI if your team already works in Google Cloud or has a Gemini Code Assist license: then everything sits in one place, with the administration you already know. If you work on your own with a personal Google account, start straight away with Antigravity CLI. If you already work with Claude or ChatGPT, Claude Code and Codex are the shortest route. And if you do not want to be tied to one provider, look at OpenCode. The best harness for each model is compared in the best AI harness for each model.
How a senior team works with Gemini CLI
A coding agent is fast. Whether the result is also good depends on how you use it. This works.
A good GEMINI.md
List the test commands, the folder structure and what is off limits. Every mistake the agent makes twice becomes a rule. The file lives in your repo, so it comes along when you move to Antigravity CLI.
Small, well-defined tasks
One function, one bug, one screen at a time. A big job in one prompt produces big changes nobody can review properly.
Permission per action
Let it read and search freely, and have the agent ask before it deletes, pushes or runs an unknown command. Use the sandbox for work you do not trust.
Know which model you get
On a free key or the free tier you work with a Flash model. It is fast and good for small jobs, but for tricky changes across many files you want a stronger model.
Review every change
What the agent produces goes through review like a colleague's work. For security, payments and customer data, a senior developer looks extra closely.
Whichever agent you choose, it writes the code. Whoever approves the code stays responsible.
Where Gemini CLI is a weaker fit
Like every coding agent, Gemini CLI is made for people who can read code. Without that, you cannot judge whether a change is safe. If you want to build something without programming knowledge, a tool like Lovable is a better place to start.
If you would rather type every line yourself with AI as a helper, an editor like Cursor fits better. And if you work with a personal Google account, Gemini CLI has little left to offer you: Antigravity CLI is now the way in.
Conclusion: Google’s agent, with two ways in
Gemini CLI is Google’s open-source coding agent, built around the Gemini models. For teams with Gemini Code Assist, an API key or Vertex AI it remains the agent. If you work with a personal Google account, you have used Antigravity CLI since June 2026, and it takes over your GEMINI.md and settings. What produces good work is the same for both: a clearly set up project, small tasks and a review by someone who knows where things can go wrong.
Gemini CLI = Google in the terminal
An open-source agent built around the Gemini models, for teams in Google Cloud. For individuals through Antigravity CLI.
Senior developer = judgement
The task, the limits and the review. That is where quality lives, and you do not leave it to the agent.
What is Gemini CLI?
Is Gemini CLI free?
What is the difference between Gemini CLI and Antigravity CLI?
What is the difference between Gemini CLI and Claude Code?
How do I install Gemini CLI?
Is Gemini CLI safe for client code?
Build faster with AI without lowering the bar?
Our senior developers work with coding agents every day. Bring them onto your project, or have your team learn to use them well in a workshop.