The company announced on Wednesday the launch of Gemini CLI, an agentic AI tool designed to run locally from your terminal. The new tool connects Google’s Gemini AI models to local codebases, and it allows developers to make natural language requests, such as asking Gemini CLI to explain confusing sections of code, write new features, debug code, or run commands.

Gemini CLI is part of Google’s efforts to get developers using its AI models in their coding workflows. Google now offers an array of AI coding tools, such as Gemini Code Assist and its asynchronous AI coding assistant, Jules. However, Gemini CLI competes directly with other command-line AI tools such as OpenAI’s Codex CLI and Anthropic’s Claude Code — tools that tend to be easier to integrate, faster, and more efficient than other AI coding tools.

Since Google launched Gemini 2.5 Pro in April, the company’s AI models have become a favorite among developers. The popularity of Gemini 2.5 Pro has driven usage of third-party AI coding tools, such as Cursor and GitHub Copilot, which have become massive businesses. In response, Google has tried in recent months to build a direct relationship with these developers by offering in-house products.

While most people will use Gemini CLI for coding, the company says it designed the tool to handle other tasks as well. Developers can tap Gemini CLI to create videos with Google’s Veo 3 model, generate research reports with the company’s Deep Research agent, or access real-time information through Google Search. Google also says Gemini CLI can connect to MCP servers, allowing developers to connect to external databases.

To encourage adoption, Google is also open sourcing Gemini CLI under the Apache 2.0 license, which is typically considered one of the most permissive. The company says it expects a network of developers to contribute to the project on GitHub.

Google is also offering generous usage limits to spur adoption of Gemini CLI. Free users can make 60 model requests per minute and 1,000 requests per day, which the company says is roughly double the average number of requests developers made when using the tool.

While AI coding tools are rising rapidly in popularity, using them comes with risks. According to a 2024 survey from Stack Overflow, just 43% of developers trust the accuracy of AI tools. Several studies have shown that code-generating AI models can occasionally introduce errors or fail to fix security vulnerabilities.