Gemini CLI

What it is

Gemini CLI is an open-source AI agent from Google that brings the Gemini models into the terminal. Google released it on 25 June 2025 under the Apache 2.0 license^[1]. The terminal is the text-based interface where developers type commands, build software, read logs and manage files; it has been one of the most basic working environments in software development for decades. Putting a model there means that asking a question in plain language, letting something act on the answer, and running ordinary commands can all happen in the same window. An "agent" in this context is not a one-question-one-answer chatbot: it is a program set up to take an instruction and work through the steps needed to carry it out. Apache 2.0 is a permissive open-source license, meaning the source code can be read, modified and redistributed.

Free access

What drew the most attention at launch was the pricing. Signing in with a personal Google account gives a free Gemini Code Assist license, with access to Gemini 2.5 Pro and its 1-million-token context window, at up to 60 model requests per minute and 1,000 requests per day^[1]. Google described this as "the industry's largest allowance" for a free tier^[1]. The numbers mean different things. A context window is the amount of text a model can hold in view at one time, so 1 million tokens is a very large working set, on the order of many long books. The per-minute and per-day request limits say how often the agent can call the model; they decide whether a long working session stays practical for someone who is not paying. For individual developers and students, a free tier of that size removed most of the cost question from simply trying an agent in daily work.

What it can do

Gemini CLI supports the Model Context Protocol (MCP) and can ground prompts with Google Search^[1]. MCP is a common standard for connecting AI applications to outside tools and data sources, so support for it means the agent can be plugged into services that were not built into it from the start. Grounding a prompt with Google Search means the model's answer can be tied to search results rather than produced by the model alone, a common way of reducing answers that are simply made up. The tool is not a separate line of work inside Google: it shares technology with the agent mode of Gemini Code Assist^[1].

Changes over time

The tool kept pace with new models. Gemini 3 Pro was made available in Gemini CLI on 18 November 2025^[2]. Its run as Google's consumer-facing terminal agent was shorter than its development history might suggest: in May 2026 Google released the Antigravity CLI, which replaced Gemini CLI for consumers^[3]. In less than a year, the terminal agent had gone from a new release with a large free allowance to a tool that a newer command-line agent took over for ordinary users.