Google has taken a major step toward integrating AI into developers' daily workflows with the release of Gemini CLI, an open-source AI agent that brings powerful language model capabilities directly to the terminal environment.
The new tool, released under the Apache 2.0 license, provides developers with immediate access to Google's Gemini 2.5 Pro model through a lightweight command-line interface. While primarily designed for coding tasks, Gemini CLI extends beyond code generation to support content creation, problem-solving, deep research, and task management—all without leaving the terminal.
"For developers, the command line interface isn't just a tool; it's home," Google stated in its announcement. "The terminal's efficiency, ubiquity and portability make it the go-to utility for getting work done."
Developers can access Gemini CLI free of charge with a personal Google account, receiving generous usage limits of 60 model requests per minute and 1,000 requests per day. For more control or higher-volume use cases, users can alternatively configure the tool with API keys from Google AI Studio or Vertex AI.
The fully open-source nature of Gemini CLI allows developers to inspect the code, understand its security implications, and contribute improvements. Google has designed the tool to be highly extensible, supporting emerging standards like the Model Context Protocol (MCP) and customizable system prompts via GEMINI.md files.
Gemini CLI also integrates with Google's AI coding assistant, Gemini Code Assist, enabling developers to seamlessly transition between terminal-based and IDE-based AI assistance using the same model backbone.
Alongside the CLI announcement, Google shared advancements in robotics applications powered by Gemini 2.5. The company's Gemini Robotics On-Device model demonstrates general-purpose dexterity and fast task adaptation, optimized to run efficiently on robotic hardware. This model operates independently of network connectivity, making it suitable for latency-sensitive applications and environments with intermittent connectivity.
The robotics model follows natural language instructions and demonstrates dexterity for various tasks, including unzipping bags, folding clothes, and assembling products. It represents Google DeepMind's first vision-language-action (VLA) model available for fine-tuning, further expanding Gemini's capabilities beyond digital environments into physical embodied intelligence.