Google announced its open-source Gemini CLI today, providing natural language command execution within developer terminals, powered by Google’s Gemini Pro 2.5.
The Gemini CLI offers a free usage tier, which includes 60 model requests per minute and a daily limit of 1,000 requests. Google established this 1,000-request limit by first assessing the usage patterns of its internal developers and subsequently doubling that observed frequency.
AI on your machine, without using your resources
The Gemini CLI functions as a local agent and incorporates built-in security protocols. Each command executed through the system necessitates explicit user confirmation, providing options to “allow once,” “always allow,” or deny specific operations. The security architecture includes multiple protection layers, such as native macOS Seatbelt support for sandboxing, the ability to run the agent within Docker or Podman containers, and the option to route network traffic through proxies for inspection. The open-source nature of the tool, governed by the Apache 2.0 license, enables comprehensive code auditing.
Ryan J. Salva, Google’s senior director for product management, stated that for the “vast majority of developers, Gemini CLI will be completely free of charge,” emphasizing the intent to avoid situations where developers monitor token usage closely. Taylor Mullen, a Google senior staff software engineer, anticipates broader adoption of Gemini CLI due to its cost-free nature, noting that competing tools from OpenAI and Anthropic, specifically OpenAI Codex and Claude Code, typically incur costs, which may limit their use for every task. Mullen indicated that the ability to apply the tool to “literally anything and everything means it’s woven into the fabric of so much more of your workflow.”
Gemini CLI operates under the Apache 2.0 license, making it an open-source tool. Its extensibility framework centers around the Model Context Protocol (MCP) standard, which facilitates the integration of external services and the addition of new functionalities.
Image and video generation plus MCP support
Google demonstrated this extensibility through an example of integrating Gemini CLI with Google’s creative AI tools, including Imagen APIs for image generation and Veo technology for video animation. The extensibility model comprises three distinct layers: built-in MCP server support, bundled extensions combining MCP servers with configuration files, and custom Gemini.md files for project-specific customization. This layered architecture supports both individual developer customization and standardized team workflows.
Accessing the Gemini CLI requires only a Google login, eliminating the need for an API key or credit card for most users. However, enterprises requiring specific functionalities will incur costs. Salva clarified that a paid API key becomes necessary if an organization intends to run multiple Gemini CLI agents concurrently, or if there are explicit requirements pertaining to policy, governance, or data residency. Such a key could facilitate access to Google Vertex AI, which provides commercial access to various models, including Gemini Pro 2.5.
Salva stated that users have “complete transparency into it” and that “the tool only has access to the information that you explicitly provide in a prompt or a reference file path and you decide what context to share with the model on a prompt by prompt by prompt basis.” While Gemini CLI operates as a local agent, it accesses the Gemini Pro 2.5 model remotely from the cloud. Google does not currently provide support for running the model locally, despite acknowledging that a subset of tasks could potentially be performed with a local model. Salva emphasized the approach to make tools accessible regardless of financial resources, stating, “It doesn’t matter if you’ve got dust or dollars, whether you’re a student, hobbyist, a freelancer or a developer at a very well funded company, you should have access to the same tools.”
