LLM powered development for IntelliJ
An IntelliJ plugin that adds ghost-text code completion backed by a language model, using the Hugging Face Inference API or any compatible HTTP endpoint, with the prompt automatically sized to the model's context window.
llm-intellij is an IntelliJ plugin for LLM-assisted development, using a companion language server as its backend. Its main feature is ghost-text code completion in the style of Copilot: as a developer types, requests for code generation are sent over HTTP to a configured model and the suggested completion is shown inline.
The plugin can call the Hugging Face Inference API or a self-hosted HTTP endpoint that follows the same interface, is assumed to work with any code-generation model, and ships example configurations for models such as Starcoder and CodeLlama, including their fill-in-the-middle tokens and context window sizes. Whatever endpoint is used, the prompt sent to the model is automatically sized to fit within that context window using a tokenizer.