How to Use Gemini CLI

Gemini CLI is Google’s open-source terminal agent that puts Gemini’s 1M-token context window behind a single `gemini` command.

Gemini CLI is Google’s answer to terminal-first AI tools like Claude Code and Codex CLI. You install it once, sign in with a personal Google account, and get access to Gemini 2.5 Pro with a generous free tier of requests per day and minute. It reads files, runs shell commands, edits code, and can pull in web search and the full 1M-token context window for questions that span a whole repo or long documents.

What it is

Gemini CLI is an Apache-2.0 project from Google on GitHub. It ships as an npm package (gemini-cli) and runs on Node. Authentication is either a personal Google sign-in for the free tier, or an API key via Google AI Studio / Vertex AI for higher limits and team usage. It follows the Model Context Protocol, so MCP servers plug in as tools.

Install

# Node 20+
npm install -g @google/gemini-cli

# first run triggers login
gemini

First session

Run `gemini` in any project directory. The REPL opens, reads a GEMINI.md file if present, and waits for a task.

$ cd my-repo && gemini
> Summarize how auth is wired in this repo
→ grep, read files, cite paths, answer

Everyday workflows

• Codebase questions — point it at a large repo and ask architectural questions; 1M context shines here.
• Scripted refactors — let it edit files in place, review the diff, commit.
• Research — built-in Google Search grounds answers in live links instead of stale training data.

Gotchas and tips

Free-tier rate limits are per-minute and per-day, so long tool-heavy tasks can stall. If you hit the ceiling, switch to an API key or drop to Gemini 2.5 Flash for the cheap iterations. The CLI respects a .gitignore-style pattern file to keep sensitive paths out of context.

Context compaction runs automatically on long sessions, but you can force it with /compress. Drop a GEMINI.md at the project root with conventions, test commands, and taboo files — it is loaded on every session and is the easiest way to get consistent output.

Who it’s for

Developers who want a free, open-source terminal agent with a very large context window, and who are fine living inside Google’s model family. Start on a read-only task to get a feel for the tool-use loop before letting it edit files.