This plugin provides access to Google's Gemini family of large language models through the llm CLI tool. It enables users to interact with various Gemini models, leverage their multi-modal capabilities, and utilize features like code execution and Google Search grounding.

How It Works

The plugin integrates with the llm command-line interface, allowing users to specify Gemini models via the -m flag. It handles API key management and configuration, supporting both direct key setting and environment variables. The plugin facilitates multi-modal inputs (images, audio, video) by accepting file paths or URLs as arguments. Advanced features like JSON output formatting, code execution, and Google Search grounding are enabled through specific command-line options.

Quick Start & Requirements

• Install via llm install llm-gemini.
• Requires an API key for Google's Gemini models.
• API key can be set via llm keys set gemini <key> or the LLM_GEMINI_KEY environment variable.
• Supports various Gemini models, including gemini-2.0-flash, gemini-2.5-flash-preview-04-17, and embedding models.
• Documentation: https://github.com/simonw/llm-gemini

Highlighted Details

• Supports multi-modal inputs (images, audio, video).
• Enables code execution within a secure sandbox.
• Integrates Google Search grounding for up-to-date information.
• Provides access to multiple Gemini models and embedding models.
• Supports interactive chat sessions.

Maintenance & Community

• Developed by Simon Willison.
• Project repository: https://github.com/simonw/llm-gemini

Licensing & Compatibility

• License: MIT License.
• Compatible with the llm CLI tool.

Limitations & Caveats

• Experimental Gemini models may have unstable behavior or change without notice.
• Usage of certain features like Google Search grounding may have additional requirements or costs as per Google's documentation.