Turbopilot is a self-hosted code completion engine that leverages llama.cpp to serve large language models for inline code generation. It supports multiple open code generation models including Salesforce Codegen 6B, Stability AI StableCode 3B, and WizardCoder, all runnable within 4GiB of RAM. The project functions as a local alternative to GitHub Copilot by exposing a completion API that can be integrated with IDE plugins.