Turtle-Gemma is a beginner-friendly project that uses Google's Gemma model and a headless turtle graphics engine to visually demonstrate how AI agents and tool-calling work. By speaking commands like 'draw a red star,' the AI reasons through sequential tool calls (move_turtle, turn_turtle, set_pen_color) and renders the result on a canvas. This makes the abstract concept of agentic workflows tangible — AI mistakes appear as lopsided drawings rather than silent errors, offering an intuitive way to understand LLM reasoning.
Table of contents
🐢 Speaking into Canvas🛠️ Gluing It All Together💭 Demystifying "Tool Calling" by Watching the AI Think😋 Embracing the Happy Accidents🏖️ Go Play in the SandboxSort: