What is Gemini CLI? Install, sign in and use it in 2026

What is Gemini CLI?

Gemini CLI is an open-source AI agent from Google that runs in your terminal. You start it with the command gemini inside a project directory, describe a task in plain English, and it can query and edit a codebase, run shell commands, debug problems and automate chores such as handling a complex rebase. Its GitHub repository says it can also generate apps from PDFs, images or sketches using multimodal input, and connect extra tools through MCP servers.

It launched in 2025 with free access for individuals, and that history is why most search results still call it free. The next section explains why that is no longer the whole story.

Is Gemini CLI still available, and is it still free?

It is still available, but not on the free tier. Google's announcement on its developers blog, dated May 19, 2026, says that on June 18, 2026 Gemini CLI and the Gemini Code Assist IDE extensions stop serving requests for Google AI Pro and Ultra subscribers and for people using them free of charge through Gemini Code Assist for individuals. Google moved those users to Antigravity CLI, a new terminal tool in its Antigravity platform.

The repository README now opens with a status notice that says the same thing. Gemini CLI is designed for Gemini Code Assist Standard and Enterprise licenses and Code Assist for GitHub on Google Cloud, plus legacy paid API and Vertex AI keys. It also says Gemini CLI is not tuned for the Gemini 4 series and should not be used to benchmark Gemini models.

How do you install Gemini CLI?

The installation guide lists these requirements: Node.js 20.0.0 or later, macOS 15+, Windows 11 24H2+ or Ubuntu 20.04+, a Bash, Zsh or PowerShell shell, an internet connection, and 4 GB or more of RAM for casual use (16 GB or more for long sessions on large codebases).

Then run gemini in the folder of your project. Releases follow three channels: stable (the latest tag, published weekly on Tuesdays at 20:00 UTC), preview (published weekly on Tuesdays at 23:59 UTC and not fully vetted) and nightly (a daily build from the main branch). Install a channel with npm install -g @google/gemini-cli@preview, @latest or @nightly.

How do you sign in to Gemini CLI?

Gemini CLI offers three ways to authenticate, described in the authentication guide. Run gemini, pick the method when prompted, or change it later with the /auth command.

# 1. Create a key at https://aistudio.google.com/apikey
# 2. Put it in your environment
export GEMINI_API_KEY="YOUR_API_KEY"

# 3. Start the CLI in your project and choose the API key option
cd my-project
gemini

Treat API keys as secrets. Google's docs warn that a leaked key can be misused under your account. If you already exported GOOGLE_API_KEY or GEMINI_API_KEY and want to use Vertex AI with application default credentials, unset them first. To keep variables between terminal sessions, add them to your shell profile. For automated jobs, the docs recommend a Gemini API key or Vertex AI credentials.

What is GEMINI.md and how do you write one?

GEMINI.md is a context file that gives the agent standing instructions, so you do not repeat them in every prompt. Gemini CLI loads these files and sends their combined content with each prompt, per the context file documentation. It looks in three places.

1. Global: ~/.gemini/GEMINI.md, for defaults across all your projects.
2. Workspace: GEMINI.md files in your workspace directories and their parent directories.
3. Just-in-time: when a tool reads a file or directory, the CLI also scans that directory and its ancestors, up to a trusted root, for more GEMINI.md files.

Run /init and Gemini CLI analyzes the current directory and generates a tailored GEMINI.md for you. Use /memory show to print the full combined context, /memory list to see which files are loaded and /memory refresh to reload after you edit one. You can split a long file with imports, writing @./components/instructions.md on its own line, and you can rename the file through the context.fileName setting in settings.json. For example, setting it to a list that includes AGENTS.md makes the CLI read that file too. See the AGENTS.md guide for the cross-tool version of this idea.

# Project: [name]

## What this is
[One sentence: who it is for and the one thing it does.]

## Stack
- [Language and framework]
- [Database or storage]
- Run locally: [command]. Run tests: [command].

## Rules
- Make the smallest change that solves the task. Do not refactor unrelated code.
- Follow the existing code style and naming.
- Ask before adding a new dependency.
- Never print or commit secrets. Read them from environment variables.
- After a change, run the tests and tell me the result.

## Out of scope
- [Things the agent must not touch, such as the billing code or the migrations folder]

What are the core Gemini CLI commands?

Inside a session, three prefixes control the tool. Slash commands (/) manage the session and settings. The at sign (@) injects a file or directory into your prompt, as in @src/ Summarize the code in this directory. The exclamation mark (!) runs a shell command, and typing ! alone toggles shell mode. The command reference lists them all. These are the ones worth learning first.

Outside a session, a few flags matter. Use gemini -p "your prompt" for a one-shot answer in scripts, add --output-format json for structured output or stream-json for newline-delimited events, gemini -m to pick a model, gemini --include-directories ../lib,../docs to add folders, and gemini --sandbox to run inside a container.

What does a first session look like?

1. Create or open a project folder and run git init, so you can see and undo every change.
2. Run gemini, choose an authentication method, and run /init to generate GEMINI.md. Edit it down to the rules you actually care about.
3. Switch to Plan Mode with /plan and ask for a plan before any edits.
4. Approve the plan, let it build one small piece, and run your tests.
5. Commit, then repeat. If something goes wrong, use /restore or git to go back.

Plan, then build, a small command-line tool: [one-sentence description].

Version one does only this: [input] in, [output] out.
Language: [language]. No new dependencies without asking.

First, list the files you will create and the tests you will write, and stop. Do not edit anything until I say go.

A small tool is the right size to learn on. The board has plenty of candidates, such as Explain this cron expression in plain English, and the reverse, posted under developer tools. Browse more in developer tools ideas.

How does Gemini CLI relate to Antigravity CLI?

Google describes Antigravity CLI as the single product it is putting its effort into. It is built in Go, runs on the same agent engine as the Antigravity 2.0 desktop app, and runs multiple agents in the background. Google says it keeps Agent Skills, Hooks, Subagents and Extensions (now called plugins), though without one-to-one feature parity at launch. The command is agy.

If you move, Google's migration guide says agy detects legacy Gemini CLI settings on first launch, and agy plugin import gemini converts extensions to plugins. Antigravity CLI keeps reading your existing GEMINI.md and AGENTS.md files and the global ~/.gemini/GEMINI.md, so your instructions carry over. Workspace skills move from .gemini/skills/ to .agents/skills/, and MCP servers move out of settings.json into a separate mcp_config.json file.

For the head-to-head with Anthropic's agent, see Claude Code vs Gemini CLI. For another vendor's terminal agent, see Codex CLI.