Gemini CLI by Google

What Is Gemini CLI? Google's Open-Source Terminal AI Agent (2026)

Last verified: June 16, 2026  ·  Format: Breakdown

Gemini CLI is Google's open-source AI agent for your terminal. Instead of returning text for you to copy out of a chat window, it acts on your codebase directly: it reads and writes files, runs shell commands, and works through multi-step tasks while you approve each change. So if you have been wondering what is Gemini CLI in practical terms, the short answer is that it puts Google's Gemini models behind an agent that lives where you already work, the command line.

Picture typing "find every place we call the old billing endpoint, update them to the new one, and run the tests." The agent locates the relevant files, makes the edits, runs your test suite, and reports back, all without you leaving the shell. The free-tier numbers cited below are reported by Google and were checked on June 16, 2026; always confirm the current limits in the official Gemini CLI repository and docs before you build on them.

What Is Gemini CLI

Gemini CLI is an open-source AI agent from Google, released under the Apache 2.0 license, that brings the Gemini models into your terminal. Rather than returning a snippet of text for you to copy, it acts: it reads and writes files, runs shell commands, fetches web pages, and works through multi-step tasks on its own, pausing for you to approve changes.

The distinction worth drawing is between a chatbot and an agent. A chatbot answers a question. Gemini CLI is built to take a task. You describe what you want in plain language, and the agent plans the steps, uses tools to carry them out, observes the results, and adjusts. Because the project is fully open source, the entire codebase is on GitHub for anyone to inspect, fork, or extend, which is unusual for a flagship coding agent and matters if you care about auditability or self-hosting your tooling.

It runs on macOS, Linux, and Windows, and it needs Node.js 20 or newer. You can run it on demand with npx without installing anything, install it globally through npm, Homebrew, MacPorts, or Anaconda, or use it with no setup at all inside Google Cloud Shell. For where Gemini CLI sits among the broader Google AI ecosystem, see the Google Gemini hub, and for the wider landscape of developer AI tools, the AI Tools Hub.

How the ReAct Loop Works

The engine behind Gemini CLI is a ReAct loop, short for reason and act. Instead of producing one big answer in a single shot, the agent works in a cycle: it reasons about the next step, takes an action by calling a tool, observes what the tool returns, then reasons again with that new information. It repeats this until the task is done or it needs you.

In practice, a single request like "add input validation to the signup handler and write a test" might unfold as several turns: search the codebase to locate the handler, read the file, propose an edit, write the change, run the test command in the shell, read the failure output, and patch again. Because each action feeds the next round of reasoning, the agent can recover from its own mistakes rather than stopping at the first error. You stay in the review seat, approving file writes and commands as they happen.

This loop is also what makes the built-in tools matter so much. Reasoning alone cannot change your files or run your tests. The tools are the "act" half of the loop, and the next section walks through what ships in the box.

Core Features at a Glance

The ReAct loop is the engine, but Gemini CLI is several connected pieces. Here is what each one does and why it matters day to day.

Built-in tools

Out of the box, the agent can perform file operations (read, write, and edit code), run shell commands (run tests, install packages, use git), fetch web pages to pull in external content, and use Google Search grounding so its answers can be anchored to current results rather than only its training data. These are the actions the reasoning loop draws on, and they are what turn it from a code generator into an agent that can actually move work forward.

MCP support

Gemini CLI speaks the Model Context Protocol, an open standard for connecting AI agents to external tools and data sources. You configure MCP servers in a settings file (under your ~/.gemini/ directory), which lets the agent reach systems beyond the built-in toolset, such as a database, an issue tracker, or an internal API, through one shared interface rather than bespoke integrations.

GEMINI.md, checkpointing, and caching

You can place a GEMINI.md file in a project to give the agent standing context, such as conventions, architecture notes, or commands to prefer, so it does not have to relearn your project every session. Conversation checkpointing lets you save and resume a working session, and token caching reduces repeated work on context the agent has already processed. Together these keep long tasks coherent and cut the friction of starting over.

Scripting and the GitHub Action

Beyond the interactive prompt, Gemini CLI runs non-interactively for scripting and automation, including structured output via --output-format json or a streaming JSON mode, which makes it easy to wire into pipelines. There is also an official GitHub Action that brings the agent into your repository for tasks like pull-request review and issue triage, and you can mention @gemini-cli to invoke it on a thread.

Which Models It Runs and the Context Window

Gemini CLI runs on Google's Gemini 3 model family, using a mix of Flash and Pro tiers depending on the task, and it offers a context window of up to 1M tokens. A context that large is what lets the agent reason across a big slice of a codebase at once, holding many files, logs, or docs in view rather than working one fragment at a time.

One note on version history, because it affects older write-ups you may find. Earlier previews of the tool referenced Gemini 2.5 Pro; current docs describe the Gemini 3 family. Treat the specifics as a snapshot taken on June 16, 2026 and confirm the live model details in the project documentation, since the model lineup changes quickly.

Authentication and the Free Tier

Gemini CLI gives you several ways to authenticate, and which one you pick shapes both your limits and your billing. The headline is that there is a genuinely usable free tier: signing in with a personal Google account gives you a free allowance that Google reports at 60 requests per minute and 1,000 requests per day, with access to the Gemini 3 models and the 1M-token context, and no API key to manage. These numbers are vendor-reported and were checked on June 16, 2026; Google frames the allowance as generous but subject to change, so confirm the current figures before you rely on them.

The other paths suit different needs. A Gemini API key from Google AI Studio lets you stay on a free daily allowance and then move to usage-based billing when you need more; note that two docs pages have disagreed on the exact free API-key limit (one citing 1,000 requests/day on Gemini 3, an older page citing 100/day on Gemini 2.5 Pro), so the figure depends on the model version and you should check the current docs. Vertex AI is the route for enterprise teams and production workloads, offering higher rate limits through a Google Cloud billing account. And if your organization already uses Gemini Code Assist, its Individual, Standard, or Enterprise quotas are shared between the CLI and Code Assist.

How It Connects to Gemini Code Assist

If you prefer an editor to a terminal, Gemini CLI still reaches you. Google has stated that the agent mode in Gemini Code Assist for VS Code is powered by Gemini CLI: the editor experience runs on a subset of the CLI's functionality. That means the same reason-and-act engine that drives the terminal agent also drives the agentic features inside the IDE.

The practical upshot lands in two places. First, skills and habits transfer: what you learn about prompting and reviewing in the terminal applies in the editor. Second, as noted above, the quotas are shared, so a team on a Code Assist plan draws from one pool whether they work in VS Code or at the command line. If you are choosing between the two surfaces, it is less an either-or and more a question of where you prefer to sit for a given task.

Who Gemini CLI Is For

Gemini CLI suits a wide range of developers, but the right auth path and the value shift by use case. Here is how the main groups line up.

Honest Limitations

Gemini CLI is a capable tool, and the points below are not reasons to avoid it. They are reasons to use it with clear eyes.

Frequently Asked Questions