Commodore 64 toolchain using agents and Google Gemini 3 to develop assembler code for the Commodore 64. This project demonstrates a workflow where AI models can write, debug, and even playtest 6510 assembly games by providing them with a structured feedback loop involving the VICE emulator and Python-based visual verification.
# C64AIToolChain **C64AIToolChain** is a Commodore 64 development toolchain designed as a **creative benchmark for AI agents**. It uses AI models inside **Visual Studio Code with GitHub Copilot** (agent mode) to develop C and assembly games for the Commodore 64. **Tested with:** **Google Gemini 3** · **Claude Opus 4.6** ### How It Works as a Benchmark The agent is given access to the workspace via GitHub Copilot in VS Code and asked to: 1. **Analyze** the project structure, existing games, and development rules (`AGENT_RULES.md`, `AGENT_HOWTO.md`) 2. **Understand** the constraints (6502 CPU, 64KB RAM, VIC-II, SID, cc65 compiler) 3. **Generate a game** — either a classic clone or, for the creative benchmark, an entirely **new original game** by combining mechanics from existing ones This tests sustained, multi-domain autonomous problem-solving: game design, systems programming, hardware constraints, memory layout, visual debugging, and iterative refinement — all in a single unbroken session. Unlike benchmarks such as ARC-AGI (pattern recognition), SWE-bench (isolated bug fixes), or HumanEval (function-level generation), this measures an agent's ability to **hold a complex constrained system in context and ship a working product**. > **All `.prg` files are included pre-compiled** — load them directly in VICE without needing cc65. --- ## 🎮 Game Showcase ### ☄️ METEOR STORM — *AI Original (Claude Opus 4.6)* <p align="center"> <img src="meteor/meteor.gif" alt="Meteor Storm Demo" width="480"> </p> An **original game** (not a clone) created entirely by Claude Opus 4.6 via GitHub Copilot. The agent analyzed the existing codebase and designed something new by combining mechanics from Asteroids (splitting meteors), Space Invaders (destructible shields), and Arkanoid (power-up drops). **1,581 lines of C** · 14 custom characters · 4 hardware sprites · 3-voice SID sound · parallax starfield · combo scoring · UFO bonus · demo AI · progressive wave difficulty. The
Google's AI-powered research notebook that ingests your documents and becomes an expert on your content. Generates audio overviews, study guides, FAQs, and interactive discussions from uploaded sources.
Google DeepMind's experimental AI agent that can navigate websites, fill forms, and complete multi-step browser tasks autonomously. Uses Gemini's multimodal understanding to interact with web interfaces.
Google DeepMind's universal AI assistant prototype that can see, hear, and respond in real-time through your device camera and microphone. Demonstrates the future of multimodal AI interaction.
Google Cloud's enterprise platform for building, deploying, and managing AI agents powered by Gemini. Supports multi-agent orchestration, tool integration, and enterprise governance.
Gemini's agentic research capability that autonomously browses the web, synthesizes information from dozens of sources, and produces comprehensive research reports on any topic.
Interactive coding and content creation agent that generates, previews, and iterates on code, documents, and interactive applications in a side panel. Supports HTML/CSS/JS, Python, and more.