# LiteCode

A command-line AI coding agent built for small, free, or local language models with 8000-token context windows instead of large paid models.

LiteCode is a command-line AI coding agent designed to work with free-tier or local models that have small, 8000-token context windows, rather than requiring a large paid model. A user describes a code change in plain English and LiteCode executes it across the project, touching every file that needs changing in the right order, and its token budget is enforced in code before every model call rather than relying on the model to behave.

It remembers the last four actions it performed in a project and passes that memory to both its planner and its executor, so instructions such as undoing the last change or adding a related function are resolved against what it actually did. Independent file edits run in parallel, while executors run one at a time automatically when the configured provider is local, to avoid overloading it. Context maps are stored as plain Markdown, and every change shows a diff before it is written to disk.

It works with Groq, OpenRouter, Ollama, LM Studio, Gemini, and DeepSeek, and offers an interactive terminal interface with scrollable history and a diff viewer.

Website: https://github.com/razvanneculai/litecode