huggingface/llm-intellij ? reverse-engineered prompt

Build me an IntelliJ plugin that gives LLM powered code completion like Copilot, with ghost text suggestions right in the editor.

I want it to use llm ls as the backend, and let me point it either at the Hugging Face Inference API or my own HTTP endpoint. Please add a simple settings screen where I can enter an API token, choose the model or endpoint URL, and set the llm ls binary path if I want to use my own build. If the binary is missing, have the plugin download it automatically the first time it runs and cache it locally.

Make sure the prompt sent to the model always fits the model context window by using tokenizer based token counting, and support a couple of common model presets like Starcoder and CodeLlama. Keep the setup easy, works out of the box, and show clear errors if the token, endpoint, or backend is misconfigured. If you need to, look up current docs online for the API details.