intellimetrics Learning
Building Module 6 · CLI Tools Setup

Gemini CLI

What you'll learn

~15 min
  • Install Gemini CLI globally via npm
  • Authenticate with a Google account or API key
  • Run your first task and view the output
  • Understand when to choose Gemini CLI vs Claude Code
ℹCloud Equivalent

Google Cloud Shell at shell.cloud.google.com has Gemini CLI tooling pre-installed, or use Gemini Web at gemini.google.com for conversational tasks. See the Cloud Sandbox Cheat Sheet for current pricing, or set up your sandbox.

By the end of this lesson, you’ll have Gemini CLI running and understand why its large context window and generous daily allowance make it a strong everyday tool.

What is Gemini CLI?

Gemini CLI is Google’s open-source command-line AI tool. It has quickly become a very popular AI CLI tool, with tens of thousands of GitHub stars. It uses Google’s Gemini models and offers a generous usage allowance.

Key features:

  • Up to 1,000 requests per day on the free Google-account tier (60 req/min, full model family)
  • Massive context: 1 million tokens — it can read enormous codebases
  • Plan mode: Built-in planning that lets you review and edit a strategy before execution
  • Project context: GEMINI.md files give the model persistent project instructions (like Claude Code’s CLAUDE.md)
  • Open source: You can see exactly how it works, and the community contributes improvements
  • MCP extensible: Can connect to external tools and services
  • Multiple models: Access to Gemini 2.5 Pro, 3.1 Pro Preview, Flash, and more

Installation

Step 1: Install Gemini CLI

Terminal window
npm install -g @google/gemini-cli

Verify:

Terminal window
gemini --version
ℹWindows native works identically

Gemini CLI runs on Windows native (PowerShell, CMD, or Git Bash) with no differences — it only needs Node.js. The install command is the same whether you’re in WSL or native Windows.

Step 2: Authenticate

Run Gemini CLI:

Terminal window
gemini

It will open your browser to authenticate with your Google account. Sign in, and you’re done.

💡Authenticating over SSH or on a remote server

If you’re on a remote server without a browser, Gemini CLI supports device flow authentication — it shows a URL and a code. Open the URL on your local machine’s browser, enter the code, and the remote tool authenticates. Alternatively, you can set a Gemini API key as an environment variable: export GEMINI_API_KEY="your-key". Get one from aistudio.google.com. If you already use Google Cloud, you can also authenticate via gcloud auth — see the official Gemini CLI docs for details.

ℹInstall paths may change

Gemini CLI is under active development. If the install command above does not work, check the official repository for the latest instructions.

💡The daily allowance is generous

The free Google-account tier gives you 1,000 requests per day at 60 requests per minute, with access to the full Gemini model family. If you authenticate with an unpaid API key instead, you get a smaller allowance — 250 requests per day with Flash models only. For learning and real projects, the Google account tier is the way to go.

Your first interaction

Step 1: Create a project folder

Terminal window
mkdir hello-gemini && cd hello-gemini

Step 2: Start Gemini CLI

Terminal window
gemini

Step 3: Give it a task

Create a simple HTML page with a counter. It should have a dark
background, a large number in the center, and + / - buttons to
increment and decrement. Use vanilla JavaScript — no frameworks.

Watch as Gemini:

  1. Plans the approach
  2. Creates the file(s)
  3. Writes the HTML, CSS, and JavaScript
  4. Reports what it did

Step 4: Check the result

Terminal window
ls
cat index.html
Terminal window
open index.html

From WSL:

Terminal window
explorer.exe index.html

Or right-click the file in VS Code’s Explorer panel.

Terminal window
xdg-open index.html

On a remote server, use VS Code’s file explorer or download the file.

Gemini CLI vs Claude Code

Both are excellent tools. Here’s how they compare:

AspectGemini CLIClaude Code
AccessGoogle account (generous free tier)API key or subscription
Context window1M tokensLarge (varies by model)
Open sourceYesNo
Project contextGEMINI.mdCLAUDE.md
Best forLarge codebases, budget-conscious workComplex reasoning, nuanced instructions
SetupGoogle accountAPI key or subscription

Context windows and model capabilities change frequently — check each tool’s official docs for current limits. Many developers use both — Gemini CLI for everyday tasks and Claude Code for complex, multi-step projects where reasoning quality matters most.

🧬In Your Field: Biotechclick to expand

Gemini CLI’s large context window is especially useful for bioinformatics. You can feed it an entire sequencing pipeline — dozens of scripts, config files, and documentation — and ask questions like “Trace the data flow from raw FASTQ files to the final variant call set” or “Which steps in this pipeline could be parallelized?” It is among the largest context windows currently available in popular CLI tools, making it ideal for understanding large, inherited analysis codebases.

🏛️In Your Field: Government / State Devclick to expand

For government and state development work, Gemini CLI’s free daily allowance is a significant advantage. When you’re working on internal tools and don’t have budget approval for API subscriptions yet, 1,000 requests per day lets you prototype and iterate without procurement paperwork. It’s also useful for analyzing large policy documents or legislative text — that million-token window can hold hundreds of pages of regulations at once.

Useful commands

In-session commands (type these while Gemini CLI is running):

CommandWhat it does
/helpShow available commands
/planEnter Plan Mode — Gemini outlines a strategy before executing
/modelSwitch between Gemini models
/memory showView the memory Gemini has stored about your project
/memory reloadReload memory and GEMINI.md context files
/clearClear conversation history
Shift+TabToggle Plan Mode on/off
Ctrl+CExit Gemini CLI

Startup flags (add these when launching):

FlagWhat it does
--yoloLaunch in autonomous mode (auto-enables sandbox)

Tips

  1. Take advantage of the large context. Gemini CLI’s 1M-token context window can handle very large codebases — don’t hesitate to point it at big projects.

  2. Experiment without hesitation. The free-tier daily allowance is generous — don’t worry about “wasting” requests while learning.

  3. Set up a GEMINI.md early. Even a few lines of project context (“This is a Python Flask app, we use pytest, follow PEP 8”) makes every response better.

  4. Use Plan Mode for anything non-trivial. It’s on by default, but get in the habit of reviewing the plan before letting Gemini execute. Press Ctrl+X to edit the plan externally if you want to adjust the approach.

  5. Try different models. Use /model to switch between available models — Pro for complex tasks, Flash for quick ones.

Power features

Plan Mode

Plan Mode is on by default as of v0.34.0. Gemini outlines a strategy before executing, so you can review and approve the approach. You can enter it explicitly with /plan, toggle it with Shift+Tab, or just phrase your request as a plan (“plan how to refactor the auth module”). Plans can be edited externally — press Ctrl+X to open the plan in your editor, tweak the steps, and Gemini follows your revised plan.

GEMINI.md project context

Just like Claude Code uses CLAUDE.md, Gemini CLI loads GEMINI.md files for persistent project instructions. It checks hierarchically — from ~/.gemini/GEMINI.md (global defaults) down through parent directories to your project root. Use these to set coding conventions, preferred libraries, or domain context that Gemini should always know about.

Thinking mode

Gemini CLI has a built-in thinking mode that may show planning and thought summaries depending on the model and your settings. This is enabled by default and helps you understand how it approaches a problem. For simple tasks where you do not need to see the reasoning, you can disable it in settings.

YOLO mode

Run gemini --yolo to enable autonomous mode. Gemini will execute file edits and commands without asking for confirmation. When YOLO mode is active, Gemini automatically enables sandbox isolation to contain any damage.

⚠YOLO mode still requires caution

The sandbox protects your system, but the AI can still make changes inside the sandbox that are hard to undo. Use YOLO mode for quick prototypes and learning experiments, not for work you care about keeping.

Sandbox

Gemini CLI can run inside a sandbox that isolates file changes and commands from your real system. It supports Docker, gVisor (runsc), and experimental LXC container isolation. The sandbox is auto-enabled in YOLO mode, but you can also use it manually for extra safety on any task.

Memory and checkpointing

Gemini CLI remembers context across sessions. Use /memory show to see what it has stored and /memory reload to refresh. It also supports checkpointing — session snapshots you can rewind to if an approach goes wrong, similar to save points in a game.

Model switching

Use /model mid-session to switch between models. Gemini 2.5 Pro and 3.1 Pro Preview are the most capable and best for complex, multi-step tasks. The Flash model is faster and uses less of your daily allowance — good for simple edits and quick questions. Available models may change as Google releases new versions.

Headless mode

For scripting and automation, Gemini CLI supports headless mode — run it without an interactive session, pipe input/output, and integrate it into CI pipelines or shell scripts.

Verify it works

Paste this exact command to confirm everything is set up correctly:

Terminal window
gemini --version

You should see a version number printed. If you get “command not found,” revisit the installation step above.

💡Your first real prompt

If the version check worked, try a quick test. You can pass a prompt directly as a command-line argument:

Terminal window
gemini "What is 2 + 2? Reply with just the number."

If your version does not support one-shot prompts, start an interactive session with gemini and type your prompt there. If you see a response either way, congratulations — Gemini CLI is fully working.

🔧

When Things Go Wrong

Use the Symptom → Evidence → Request pattern: describe what you see, paste the error, then ask for a fix.

Symptom
Google account authentication fails or browser doesn't open
Evidence
Error: Could not open browser for authentication or 'Authentication failed'
What to ask the AI
"Gemini CLI tried to open my browser for Google sign-in but it failed. I'm on a remote server via SSH. How do I authenticate without a browser using device flow or an API key?"
Symptom
Gemini CLI says 'quota exceeded' even though you haven't used much
Evidence
Error: 429 Resource has been exhausted (e.g. check quota)
What to ask the AI
"I'm getting a quota error from Gemini CLI but I've only made a few requests today. Could this be a rate limit rather than a daily limit? How do I check my remaining quota?"
Symptom
npm install succeeds but 'gemini' command not found
Evidence
bash: gemini: command not found
What to ask the AI
"I installed @google/gemini-cli globally with npm but my terminal can't find the 'gemini' command. How do I check if npm's global bin directory is in my PATH?"
💬Another tool in your arsenal

Each CLI tool you set up widens what you can do. Gemini CLI’s free tier means you can experiment without worrying about cost — and the skills transfer directly to any AI tool. You’re not just learning Gemini. You’re learning orchestration. The specific commands will change; the ability to direct an AI agent to do your work faster will not.

Key Takeaways

  • Install with npm install -g @google/gemini-cli — a Google account gets you 1,000 requests/day with the full model family
  • Plan Mode is on by default — review the strategy before execution, edit plans with Ctrl+X
  • GEMINI.md gives persistent project context — hierarchical loading from global to project directory
  • The 1M-token context window lets Gemini CLI read more of your project than most other tools
  • Use /model to switch models — Pro for complex tasks, Flash for quick ones
  • Pair it with Claude Code — use Gemini for broad exploration and Claude for precision reasoning