Cloud Parts

Gemini CLI

Google’s open-source terminal agent for Gemini models; since 18 June 2026 it serves only paid API keys and business licences, as Google moved free and consumer users to Antigravity CLI.

What it is
AI coding agents (terminal agents)
Type
Commercial product
Made by
Google
License
Apache 2.0
Facts checked
9 October 2026
Website Source code

What it is

Gemini CLI is a Node.js program (npm install -g @google/gemini-cli, then gemini) that gives a Gemini model built-in tools for files, shell commands, web fetching and Google Search grounding, and more through MCP servers and extensions. Instructions come from GEMINI.md context files: a global one in ~/.gemini, files in the workspace and its parent directories, and files found just in time when a tool touches a directory. The CLI concatenates them and sends them with every prompt; the file name is configurable, so a repository’s AGENTS.md can be used instead.

Safety is opt-in. Approval modes run from default (ask before acting) through auto_edit and yolo to a read-only plan mode. Sandboxing is off unless you pass -s or configure it, and then uses Seatbelt on macOS, Docker or Podman containers, gVisor or LXC. The default Seatbelt profile confines writes to the project but still allows network access. Trusted folders add a check per directory: an untrusted workspace runs in a restricted safe mode. gemini -p runs headless for scripts, and a GitHub Action runs it on pull requests and issues. New preview and stable versions ship every Tuesday, with nightly builds in between.

In May 2026 Google announced that it was unifying its efforts into Antigravity, an agent platform with a new Go-based Antigravity CLI. On 18 June 2026 Gemini CLI stopped serving requests for Google AI Pro and Ultra subscribers and for the free Gemini Code Assist for individuals. Organisations with Gemini Code Assist Standard or Enterprise licences, and users with paid API keys, keep access, and the open-source repository still releases.

Why teams pick it

  • Fully open source under Apache 2.0, so the agent loop and its tools can be read, audited and forked
  • Gemini models with a context window of 1 million tokens, and Google Search grounding built in
  • More sandbox backends than Claude Code or Codex offer: Seatbelt, Docker, Podman, gVisor and LXC
  • Weekly preview and stable releases, nightly builds and a ready-made GitHub Action

Watch out for

  • Since 18 June 2026 the free tier and Google AI Pro and Ultra plans no longer work with it; you need a Code Assist Standard or Enterprise licence or a paid API key
  • Google’s new terminal work goes into Antigravity CLI, so Gemini CLI’s long-term role is uncertain
  • The sandbox is off by default; without -s, commands run directly on your machine
  • Built for Gemini models rather than as a multi-provider agent

Alternatives

What all terminal agents have in common →
Claude CodeCommercialAnthropic’s coding agent: a terminal program, also available in IDEs, a desktop app and the cloud, that drives Claude models through a loop of reading, running and editing.OpenAI CodexCommercialOpenAI’s coding agent: an open-source CLI, written in Rust, that works in your local repository inside an OS sandbox, plus an IDE extension and Codex cloud for tasks that run on OpenAI’s machines.OpenCodeOpen sourceAn open-source (MIT) coding agent that works with 75+ model providers and local models, built as a local server with a terminal interface as one of its clients.+1 Kilo Code

How it fits with other technologies

Install, then run one sandboxed prompt · bash

npm install -g @google/gemini-cli

gemini                                  # interactive session in this directory
gemini --approval-mode plan             # read-only: explore and plan, no edits

# headless, inside a sandbox (Seatbelt on macOS, or a container)
gemini -s -p "run the test suite and summarise the failures"

See it in a real stack

No architecture uses Gemini CLI yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2025–2026

Google announced Gemini CLI on 25 June 2025 as a free, open-source AI agent for the terminal. Signing in with a personal Google account gave Gemini 2.5 Pro and its 1-million-token context window, with an allowance of 60 model requests per minute and 1,000 per day at no charge, which Google called the industry’s largest. It shared technology with Gemini Code Assist, so the same agent worked in VS Code, and it was built on emerging standards: MCP for tools and GEMINI.md for instructions.

It grew quickly. By May 2026 Google counted over 100,000 GitHub stars, 6,000 merged pull requests and hundreds of contributors. That month Google said users had outgrown the single-agent terminal and wanted many agents working together on a shared backend, and announced Antigravity CLI, written in Go and sharing a harness with the Antigravity 2.0 desktop app. Consumer and free access to Gemini CLI ended on 18 June 2026; business licences and paid API keys were unaffected.

Why it’s built this way

Open source under Apache 2.0
Google wanted developers to be able to inspect the code, verify its security implications and contribute fixes and features.
A large free allowance
Individual developers were the target, so a personal Google account came with a free Code Assist licence and limits Google expected people to rarely hit during the preview.
Standards for extension
MCP for tools, GEMINI.md for instructions and shared settings files let teams customise the agent without forking it.
Fold into Antigravity
In 2026 Google decided multi-agent work needed one backend shared by terminal and desktop, and put its energy into a single product rather than two.

Timeline

  1. 2025

    Launched as a free, open-source terminal agent with Gemini 2.5 Pro (June)

  2. 2026

    Passes 100,000 GitHub stars; Google announces Antigravity CLI as its successor for most users (May)

  3. 2026

    Stops serving free and Google AI Pro and Ultra users (18 June)

  4. 2026

    Still released weekly for business licences and paid API keys

Sources7 · facts checked 9 October 2026

Improve this page on GitHub