A fork of Google’s Gemini CLI that breaks the vendor lock-in

What it does Easy LLM CLI is a command-line agent that reads your codebase, edits files, and automates tasks like querying pull requests or handling rebases. Born as a fork of Google’s Gemini CLI, it keeps the same conversational shell interface but swaps the hard-coded backend for a pluggable one. You can point it at Gemini, OpenAI, OpenRouter, Volcengine, or any custom endpoint that mimics the OpenAI API shape, and it will use tools and MCP servers to interact with your system. It also exposes a programmatic ElcAgent class if you want to embed the agent in a Node script instead of your terminal.

The interesting bit The most useful part is not the code itself but the compatibility matrix: the maintainers actually tested fifteen different models—from GPT-4.1 to a 7B Qwen running on Ollama—across chain-of-thought, tool calling, MCP, multimodal, and token counting. It turns out many smaller local models can handle simple tasks and tool use but choke on complex jobs or multimodal inputs, which saves you from guessing. That kind of honest benchmarking is rare in a README.

Key highlights

• Forks Google’s Gemini CLI but strips the Gemini-only backend.
• Supports any OpenAI-compatible API via environment variables.
• Includes a published test matrix for 15+ models (local and hosted) across seven capability dimensions.
• Programmatic API (ElcAgent) for embedding in Node.js applications.
• Uses MCP servers to extend tool access beyond built-in shell commands.

Caveats

• Multimodal and complex-task support is model-dependent; the test matrix shows several popular local models and even some hosted ones fail or struggle in those dimensions.
• The project is a direct fork of Google’s Gemini CLI, so its core behavior and limitations are inherited rather than reimagined.

Verdict Worth a look if you like Google’s Gemini CLI interface but are locked into a different provider or running models locally. Skip it if you need a fully original agent architecture or guaranteed multimodal performance across every backend.