Install Gemini CLI

System requirements#

Before you install, check:

• Node.js 20.0.0 or later. Enforced in the package’s engines field. Older Node versions will refuse to install.
• Operating system: macOS 15+, Windows 11 24H2+, or Ubuntu 20.04+. Native Windows is supported (not just WSL).
• Shell: Bash, Zsh, or PowerShell.
• Disk: small (under 200 MB after install).

If you’re on a corporate Mac and node --version shows 18 or 19, install a current Node — brew install node or use Volta / nvm.

Install paths#

1. npm (the canonical path)#

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

This is the path the official README leads with. It works everywhere Node 20+ does — macOS, Linux, native Windows, WSL2.

2. Run without installing — npx#

npx @google/gemini-cli

Useful if you just want to try it without polluting your global npm packages.

3. Homebrew (macOS / Linuxbrew)#

brew install gemini-cli

The Homebrew formula tracks the stable channel. Run brew upgrade gemini-cli to update.

4. MacPorts (macOS only)#

sudo port install gemini-cli

Restricted enterprise environments — Anaconda fallback#

If your laptop is locked down and npm install -g fails (no admin / no global write), install Node into a Conda env first, then npm-install Gemini CLI inside it:

conda create -y -n gemini_env -c conda-forge nodejs
conda activate gemini_env
npm install -g @google/gemini-cli

The CLI then lives inside the env — conda activate gemini_env before each session.

Heads-up: there’s no standalone binary download path published for end users. The build:binary script exists in the repo (Node.js Single Executable Application support) but is an internal build tool, not a public artefact. If you want a no-Node-required experience, the Docker image (ghcr.io/google/gemini-cli:latest) is the closest thing — but it’s intended for sandboxing, not as a primary install path. See GCL.5 Pitfalls for context.

Pre-installed environments#

You don’t need to install Gemini CLI in these — it’s already there:

• Google Cloud Shell (the in-browser shell at console.cloud.google.com)
• Cloud Workstations (Google’s managed dev environments)

In both, sign in with the active Google account and you’re authenticated automatically through Application Default Credentials.

Three release channels#

Gemini CLI ships on three concurrent channels. The npm package tag controls which one you get.

Channel	Tag	Schedule	Install command
Stable	@latest (default)	Each Tuesday, 20:00 UTC	npm install -g @google/gemini-cli
Preview	@preview	Each Tuesday, 23:59 UTC	npm install -g @google/gemini-cli@preview
Nightly	@nightly	Daily, 00:00 UTC	npm install -g @google/gemini-cli@nightly

The README’s marketing line about “Powerful Gemini 3 models” refers to preview-channel access — gemini-3-pro-preview, gemini-3.1-pro-preview, gemini-3-flash-preview. The stable channel defaults to gemini-2.5-pro. If you install the default and expect Gemini 3 by default, you’ll be surprised. To switch to Gemini 3 models, either install the preview channel or switch model mid-session: gemini -m gemini-3-pro-preview.

The current stable release at the time of writing was around v0.42 (mid-May 2026); nightly was 0.44.0-nightly.20260512. Verify the exact stable number with npm show @google/gemini-cli version before depending on it.

Verifying the install#

gemini --version
gemini --help

Then start an interactive session:

gemini

You’ll be prompted to choose an authentication method. The simplest (and what most personal users want) is “Sign in with Google” — see GCL.3 Auth for the full breakdown.

Updating#

Install path	Update command
npm	npm install -g @google/gemini-cli@latest
Homebrew	brew upgrade gemini-cli
MacPorts	sudo port upgrade gemini-cli
Conda env	conda activate gemini_env && npm install -g @google/gemini-cli@latest

Uninstalling#

npm uninstall -g @google/gemini-cli
brew uninstall gemini-cli       # if installed via brew

Config and credentials live in ~/.gemini/. Remove that directory if you want a clean reset:

rm -rf ~/.gemini

What’s next#

• §GCL.3 Auth — picking a sign-in method (Google OAuth · Gemini API key · Vertex AI · Code Assist)
• §GCL.6 MCP integration — wiring Gemini CLI to filesystems, GitHub, databases via MCP servers
• §GCL.5 Pitfalls — sandbox-off-by-default, Windows icacls persistence, tool loop issue