sitdraw

How to install and use Gemini CLI: Google login and first task

Install Gemini CLI from the official docs, sign in with Google, and complete a first coding task in your repo.

To use Gemini in the terminal to read a repo and change code, finish three jobs: install Gemini CLI, sign in with a Google account, then complete a first coding task in the current project.

Follow the official get-started path instead of guessing commands. The queries are how to install Gemini CLI, Gemini CLI Google login, and the command gemini. Docs live at geminicli.com/docs/get-started/.

20+Minimum Node.js version
npmRecommended install
GoogleRecommended sign-in

Match the runtime, then pick an install path

Global npm install

Confirm node -v is 20.0.0 or newer, then run npm install -g @google/gemini-cli. That is the official default. The stable channel is the latest tag.

Homebrew or a one-off

On macOS or Linux you can brew install gemini-cli. To try once, use npx @google/gemini-cli. Do not install both npm and Homebrew copies; the PATH will fight.

MethodWho it fitsWatch for
Sign in with GoogleLocal machine with a browserPersonal accounts usually skip a project; Workspace often needs GOOGLE_CLOUD_PROJECT
Gemini API keyAI Studio users, no browserTreat the key as a secret; do not commit it
Vertex AIExisting Google Cloud projectA Cloud project and region are required

Install, sign in, first coding task

  1. 01
    Check Node and install

    Official docs require Node.js 20.0.0+ and list macOS 15+, Windows 11 24H2+, Ubuntu 20.04+. After install, run gemini and wait for the auth prompt.

  2. 02
    Sign in with Google

    Choose 1. Sign in with Google, pick the account, click Sign in. Credentials cache locally. If a work account asks for a project, set the env var before the next session.

  3. 03
    Do one verifiable job in the repo

    Start gemini at the project root. Ask for a read-only explanation of the entry file first. Then use an official-style prompt: write unit tests for an existing module, or propose one small change.

  4. 04
    Read the diff before writes

    The CLI asks before editing files or running side-effect commands. Choose Allow once; do not grant permanent access by default. Check quota with /stats model.

node -v
npm install -g @google/gemini-cli
gemini
# After login, at the repo root, for example:
# Read the entry file and explain how the app starts.
# Then write unit tests for Login.js and wait before writing files.
Does a personal Google account need a Cloud project?
Official docs say most personal accounts do not. Company, school, or Workspace accounts usually do. Personal accounts can still be asked for a project when an official exception applies.
The login page never opens. What now?
OAuth redirects the browser to localhost that the CLI listens on. Remote SSH, a headless host, or a network that blocks localhost will fail. Use a Gemini API key, or sign in on a machine that can open a browser.
How large should the first coding task be?
Pick something you can test or diff immediately. Official examples write tests for Login.js, or clone chalk and explain the source. Do not start with a whole-repo rewrite.
Does install replace a design review?
No. This is a terminal assistant. When the meeting is the same module diagram, open sitdraw.

Understand first, then approve the write

A model can narrate a change that sounds right. Without your write approval it should not touch the repo. You still run the tests.

Start the picture