Mammouth Code
Mammouth Code is Mammouth AI's terminal-based coding agent.
➡️ Get your API key — required to use Mammouth Code.
This project is open source and available on GitHub.
Installation
Run one of the following commands in your terminal.
curl -fsSL "https://code.mammouth.ai/install.sh" | bashirm "https://code.mammouth.ai/install.ps1" | iexUse in your Terminal
Simply type mammouth to run the coding agent.
Use in your IDE
For the best experience, open Mammouth Code in the integrated terminal of your IDE (VS Code, Cursor, WebStorm, etc.):
- Open the integrated terminal in your IDE
- Navigate to your project directory
- Run:
mammouth
That's it — Mammouth Code will start an interactive session right in your terminal.
Continue a previous session
There are 2 ways to continue a previous session:
- Run
mammouthas usual, then use the/sessionscommand to switch to an old session. - Run
mammouth -cto continue the last session used in the current folder.
Update
To update Mammouth Code to the latest version:
mammouth upgradeThis will fetch and install the latest version, replacing the old one while keeping your configurations and sessions intact.
Uninstall
To completely remove Mammouth Code from your system:
mammouth uninstallThis will remove Mammouth Code along with its configurations and sessions.
Token Consumption in Mammouth Code
Mammouth Code dynamically injects system prompts depending on what you ask it to do — which explains the ~10k additional tokens you may notice.
What might look like "heavy" prompts actually saves tokens on larger projects: they allow the LLM to read only the relevant files, rather than loading your entire codebase into context. That overhead looks significant when working on a single file, but the real purpose is to route the request to the right tool for the task at hand.
A simple rule of thumb:
- 📁 Use Mammouth Code if your project has 10+ files, or if you're working with a long context — which covers most real-world programming projects.
- 💬 Use the standard Mammouth interface if you're dealing with a manageable context: two or three files, ~500 lines max (e.g., trying to understand a specific function — unless you're deep into triple-inherited OOP classes 😄).
Note on Claude Haiku: This smaller model is currently used to generate session titles. This will be configurable in a future update.
