Google releases the open-source Gemini CLI

Background

Programmers work in two kinds of places. One is the editor, where code is written. The other is the terminal: the text-only window where commands are typed, programs are run, tests are executed and files are moved around. The AI assistants that arrived over the past few years mostly settled in the first place, as editor plugins or as chat windows in a browser, so reaching one meant leaving the window where the build was running. A command-line agent takes the opposite approach and lives inside the terminal itself.

The word "agent" here means a program that does more than answer a question: it takes steps on its own, such as looking at files, running commands or editing code, while the user watches and steers. Gemini CLI belongs to that category.

What Google released

On 25 June 2025 Google released Gemini CLI, described in its announcement as an open-source AI agent for the terminal^[1]. The code is published under the Apache 2.0 license, a permissive open-source license that allows a program to be used, modified and redistributed^[1]. At the time of the announcement Google described the tool as being in preview, meaning it was available to use but not yet a finished product^[1].

The free allowance

The detail that drew the most attention was the price. Individual developers signing in with a personal Google account could use Gemini 2.5 Pro at no cost, limited to 60 requests per minute and 1,000 requests per day^[1]. Google described the free allowance as "the industry's largest allowance"^[1].

Signing in with a personal Google account also granted a free Gemini Code Assist license^[1]. That license included Gemini 2.5 Pro with a context window of 1 million tokens^[1]. A context window is the amount of text a model can keep in view at one time — instructions, source files, command output, earlier turns of a conversation. A larger window lets the model work over bigger pieces of a codebase without losing the parts it read first, which is why the size of the window is one of the numbers developers compare.

Working with outside tools and the web

Gemini CLI supports MCP servers and Google Search grounding^[1]. MCP, the Model Context Protocol, is a shared interface through which a model can be connected to outside tools and data sources; supporting it means the agent can be plugged into services beyond what ships in the box. Search grounding works the other way round: rather than answering only from what the model absorbed during training, the agent can base its answer on results fetched from Google Search, which matters for questions about library versions, error messages or anything else that changes quickly.

Where it fits

Two choices define the release. The first is the license: because the client is open source, developers can read the code, see what it does with their files and their requests, and adapt it to their own workflows instead of treating it as a black box. The second is the free tier: a full model, Gemini 2.5 Pro, behind a personal account with a daily quota measured in the thousands of requests, bundled with a Code Assist license that carries the same model and the same 1-million-token window^[1]. Together they put a terminal agent within reach of individual developers and students who are not buying seats on a team plan.

The preview label is worth keeping in mind. It signals that the tool was released early, that its behaviour and its free limits could change, and that Google was inviting feedback rather than declaring the work finished.