# Gemini CLI

> Google's open-source terminal AI agent powered by Gemini models, with a 1M-token context window and built-in tools.

Website: https://geminicli.com

Gemini CLI is Google's open-source (Apache-2.0) AI agent that runs in your terminal. You install it with `npm`, `npx`, or Homebrew, point it at a project, and describe what you want in plain language. It reads and writes files, runs shell commands, fetches URLs, and grounds answers with Google Search — driven by Gemini models (the current line leads with Gemini 3) with a 1M-token context window. The same agent core also powers the Gemini Code Assist IDE extensions.

It is aimed at developers who live in the terminal and want a capable agent without leaving the shell or paying for API usage up front. Signing in with a personal Google account unlocks a generous free allowance, so you can try real agentic work before deciding whether to wire up an API key.

## Highlights

- **1M-token context** — Gemini's large context window lets the agent reason over big codebases and long docs in a single session.
- **Built-in tools** — file system operations, shell execution, web fetch, and Google Search grounding ship in the box; no plugins required for the basics.
- **MCP support** — connect Model Context Protocol servers to extend the agent with databases, APIs, and custom tooling.
- **`GEMINI.md` context files** — drop project-level instructions in `GEMINI.md` so the agent follows your conventions automatically.
- **Multimodal input** — feed it PDFs, images, and sketches, not just text.
- **Checkpointing** — enable it in `settings.json` and the agent snapshots your project before file modifications; use `/restore` to roll back any step via a shadow Git repo, without touching your real Git history.

## In an AI-assisted workflow

Gemini CLI fits the terminal-native loop: open a repo, start the agent, and let it plan and apply changes while you review. A typical first run looks like this:

```bash
npm install -g @google/gemini-cli
cd your-project
gemini
# then, at the prompt:
# > Add input validation to the signup handler and update the tests
```

It edits files on disk and runs your test or lint commands, so you review the diff rather than copy-paste from a chat window.

> [!TIP]
> Commit (or stage) your work before a large agentic task. Checkpointing is off by default — enable it in `settings.json` if you want `/restore` to undo the agent's edits — but either way a clean Git baseline makes review and rollback far easier.

## Good to know

Gemini CLI is open source under Apache-2.0 and runs on macOS, Windows, and Linux (Node.js 20+). Install via `npm`/`npx`, Homebrew, MacPorts, or Anaconda. A personal Google account has offered a free tier of 60 requests/minute and 1,000 requests/day on Gemini models; you can also bring a Gemini API key or use a Gemini Code Assist / Enterprise license.

> [!WARNING]
> Google has announced that on **June 18, 2026** it is transitioning Gemini CLI to **Antigravity CLI** and will stop serving requests for free, Google AI Pro/Ultra, and individual Gemini Code Assist users on that date. Gemini CLI keeps working through **paid Gemini API keys** and **Gemini Code Assist Standard/Enterprise** licenses, and the repo stays open source — but the free personal-account tier is ending. Note that Antigravity CLI, the replacement, is **not** published as open source, which has drawn criticism from contributors. Check the [repo](https://github.com/google-gemini/gemini-cli) and Google's developer blog for the current state before relying on the free tier.

---

_Source: https://agentscamp.com/tools/gemini-cli — Tool on AgentsCamp._