Cthulhu Code is an offline, terminal-based coding assistant that runs locally using an Ollama backend — always asking for your approval before making any changes.
A coding assistant that respects your privacy and runs entirely on your machine.
Cthulhu is an offline, terminal-based coding assistant that runs locally using an Ollama backend. It can read, search, edit, and execute code in the current project — never sending your code to any external service.
Cthulhu offers a comprehensive set of features including code explanation, feature addition, bug fixing, refactoring, and documentation writing. It operates entirely offline with local inference using Ollama backends.
All edits and shell commands require explicit approval. Cthulhu operates with transparency — you always know what it plans to do, and nothing happens without your consent.
The project follows a modular architecture with core packages including agent orchestration, Ollama backend implementation, configuration handling, terminal utilities, file-system tools, and console rendering.
Cthulhu can be used for explaining code, adding features to projects, fixing bugs, refactoring existing code, and writing documentation. It supports operations like reading and searching files, editing with diff previews, and running shell commands after user approval.
Cthulhu supports different modes — free, plan, and implement — allowing you to choose how autonomous you want it to be. All operations work within a user-in-the-loop approval workflow.
Cthulhu Code executes shell commands and modifies files on your system. As with any AI coding agent, no sandbox is safe. We strongly recommend running Cthulhu inside an isolated environment such as a Docker container, Podman pod, or virtual machine to limit its access to your host system.
The Cthulhu Code source will be published later this year on codeberg.org under the MIT License.