This project provides an LLM-powered Language Server Protocol (LSP) server designed to enhance code completion and development workflows. It targets developers seeking intelligent, context-aware code suggestions within their IDEs, abstracting the complexities of LLM interaction for IDE extension developers.

How It Works

llm-ls acts as a central hub for LLM-driven code intelligence. It leverages the current file's content to construct context-aware prompts, supporting "fill-in-the-middle" capabilities and managing tokenization to respect model context windows. The server parses code Abstract Syntax Trees (ASTs) to intelligently determine the type of completion needed (single-line, multi-line, or none).

Quick Start & Requirements

• Install: pip install llm-ls
• Prerequisites: Python 3.8+, compatible LLM backend (Hugging Face Inference API, text-generation-inference, ollama, OpenAI-compatible APIs).
• Docs: https://github.com/huggingface/llm-ls

Highlighted Details

• Supports multiple LLM backends including Hugging Face Inference API, TGI, Ollama, and OpenAI-compatible APIs.
• AST parsing for intelligent completion type determination.
• Context window management via tokenization.
• Telemetry for potential model retraining (data stored locally).

Maintenance & Community

• Actively developed by Hugging Face.
• Roadmap includes multi-file context, prompt tuning, and filtering bad suggestions.

Licensing & Compatibility

• License: Apache 2.0.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

This project is explicitly marked as a work in progress, with the potential for broken functionality. Support for Jupyter integration is currently missing.