Kuzco

Overview

Kuzco brings AI-powered code completion to Neovim while keeping everything local and private. The plugin interfaces with Ollama to run large language models on your own machine, generating function implementations based on the function signature—name, argument types, and return type. Rather than relying on cloud APIs, Kuzco leverages locally-running models, meaning your code never leaves your machine. This makes it suitable for proprietary codebases and environments without internet access, while avoiding API costs. The plugin integrates seamlessly with Neovim's completion framework. When you define a function signature, Kuzco can suggest a complete implementation that respects the types and naming conventions in your code. It understands context from the surrounding code to generate relevant suggestions.

Features

• Local LLM inference via Ollama
• Context-aware function completion
• Type-informed suggestions
• Privacy-preserving - code never leaves your machine
• Seamless Neovim integration
• Works offline

Challenges

Crafting prompts that consistently produce useful completions required iteration. The LLM needs enough context to understand what's needed, but too much context slows inference and can confuse the model. Finding the right balance of function signature, surrounding code, and instructions took experimentation.

What I Learned

Working with local LLMs taught me about prompt engineering in a practical context. The constraints of running models locally—memory limits, inference speed—forced creative solutions for keeping the experience responsive.

Future Plans

Support more completion triggers beyond function signatures. Add inline suggestions as you type. Implement caching for common patterns.