Google has recently introduced Gemini CLI, an open-source AI agent designed to operate directly within the terminal. Announced on June 25, 2025, this tool provides a new way to access the Gemini 2.5 Pro models: developers can now engage with AI using natural language right from their terminal via the shell, integrating seamlessly into their regular workflow. Google has made this tool freely available and customizable, promising that it offers much more than just code generation capabilities.
Gemini CLI, a Free AI Agent Tailored for the Terminal
In a blog post, Google emphasized that a “CLI is not just a tool; it’s a home” for many developers, which inspired the integration of a native agent. With Gemini CLI, Google aims to place AI “on the most direct path between your prompt and [its] model”. The tool is released under the Apache 2.0 license, allowing anyone to inspect, fork, or modify the code freely, a stark contrast to the more restrictive licenses typically associated with commercial AI assistants.
The project leverages the agent branch of Gemini 2.5 Pro, previously used in the Gemini Code Assist extension for Visual Studio Code. In practice, the CLI (Command Line Interface) connects to the Gemini service using an authentication token (free Code Assist license or Vertex AI/AI Studio key) and processes requests almost in real-time. Google highlights its extended context window of one million tokens, which is designed to analyze entire projects without manual segmentation.
More Than Code Generation, a Swiss Army Knife for the Shell
While its primary function is programming assistance (code explanation, writing tests, debugging, script execution), Google points out that Gemini CLI “excels at coding, but is built to do much more”. The agent can pull in Google Search to enhance its responses with real-time web sources, interact with databases or third-party services via the Model Context Protocol (MCP), and even initiate in-house multimedia generators. The official demo showcases the automatic production of a short video using Veo and Imagen.
Additionally, the CLI can be used non-interactively within a script, paving the way for automating DevOps tasks (creating reports, automated commit reviews, drafting changelogs). This versatility meets the growing demand for AI agents capable of “planning multiple steps, recovering from failures, and offering alternative solutions”, a concept already introduced with the Code Assist agent.
Generous Quotas to Attract Independent Developers
Google emphasizes that Gemini CLI primarily serves as a democratized gateway to its models. “Gemini CLI brings Gemini directly into developers’ terminals, providing individuals with unparalleled access”, the company states. A personal Google account is enough to obtain a free license and benefit from substantial quotas: 60 calls per minute and 1,000 calls per day, free of charge during the preview phase. According to the official statement, these limits are “the most generous allocation in the industry” for a free service.
For teams with heavier demands, it’s still possible to switch to a Code Assist Standard or Enterprise subscription, or use a pay-as-you-go key via Vertex AI. In doing so, Google aims to match offerings from competitors like OpenAI (Codex CLI) and Anthropic (Claude Code), while avoiding the high entry barriers often criticized in other tools.
An Extensible Architecture Aiming at Community Involvement
To accelerate adoption, Google is betting on external contributions. The GitHub repository features a plugin system based on MCP, as well as a GEMINI.md file where each user can define persistent system prompts. “We understand that the terminal is a personal space; everyone deserves the autonomy to make it unique”, the project authors write.
The installation is performed via pip or npm, depending on the chosen environment. An interactive assistant then guides the connection to the Google account. Once set up, the agent is invoked with the command gmn, accepting conversational instructions like “Explain this Bash script”, “Write a unit test for this file”, “Deploy the staging branch on GKE”, etc. Responses appear directly in the console, with options to confirm or modify the suggested actions before execution.
Google assures that no project data is stored outside the session context unless the user opts to activate continuous tracking to enrich suggestions. This aligns with the data residency principle already implemented in Code Assist, a critical point for organizations concerned about sovereignty and code confidentiality.
Similar Posts
- Google’s AI Code Agent Jules: How to Test Its Cutting-Edge Features!
- OpenAI Unveils Codex: Revolutionizing Automated Coding with ChatGPT AI Agent
- Meet Claude 4 Opus and Sonnet: Anthropic Unveils Its Latest AI Innovations
- Gemini Update: Google Supercharges API with 2.5 Flash, Pro, and New Multimodal Features
- GitHub Unveils New AI Code Agent: Discover How It Works!
Jordan Park writes in-depth reviews and editorial opinion pieces for Touch Reviews. With a background in UI/UX design, Jordan offers a unique perspective on device usability and user experience across smartphones, tablets, and mobile software.