Understanding myPip's AI Engine (LLMs)

Learn how Large Language Models (LLMs) and advanced AI agents power the MyPip game engine as well as how to use it efficiently to build your games!

🧠 What is an LLM?

LLM stands for Large Language Model. MyPip uses cutting-edge models like Anthropic's Claude 3.5 Sonnet and Claude 4 to generate your game logic, physics, state loops, and UI based entirely on your natural language prompts.

Think of an LLM as an incredibly advanced, context-aware autocomplete system for code. Having analyzed billions of lines of code and game architecture patterns, it doesn't "think" like a human, but it can instantly predict and generate the exact code needed to make your game run.

In MyPip, the LLM acts as your automated lead game developer, translating concepts like "make the player bounce higher when they collect a coin" into structured, playable mobile code.

🔒 Privacy First: MyPip never uses your proprietary game concepts, prompts, or project data to train public AI models. Your code is yours.

✍️ What’s a Prompt?

A prompt is the creative direction, mechanic, or instruction you give to the AI.

Examples of game prompts in MyPip:

• “Build an endless runner game where a neon triangle dodges incoming obstacles. Increase the game speed every 10 seconds.”

• “Create a rogue-lite card battler screen with health bars, an energy system, and a hand of 3 draggable attack cards.”

The more specific you are about your game's rules, states, and visuals, the better the engine will execute it. Check out our Game Prompting Guide to master the art of vibe coding.

🧩 What is Game Context?

Context is the "working memory" the AI uses while building your game. To write code that integrates seamlessly with what you’ve already built, the AI looks at:

• Your latest prompt

• The entire existing game code and component structure

• Previous iterative tweaks (e.g., bug fixes or asset changes)

• Active state variables (e.g., how player scores and health are tracked)

LLMs have a context window—the maximum amount of data they can process in a single look. While MyPip uses models with massive context windows, exceptionally long game-jamming sessions with hundreds of changes can push those limits, causing the AI to occasionally forget early rules.

Tip: If the AI begins losing track of your core rules, you can optimize performance by using our Reset Context tool.

🔢 What Are Tokens?

LLMs don't read raw text; they break words and code down into small structural units called tokens.

• "Jump!" = 1-2 tokens

• <Sprite type="player" velocity={y: -5} /> = ~12 tokens

Every mechanic you add, every bug you fix, and every screen the AI generates consumes tokens. This matters because token usage dictates how much memory the AI has available for your session and maps directly to your plan's resource usage.

📊 Estimated Token Footprint in MyPip (How to tokens convert to Build Credits?)

Game ElementEstimated Token UsageSimple Game Screen (e.g., Main Menu or Splash Screen)100–300 tokensBasic Game Loop (e.g., Clicker or Trivia mechanic)600–1,200 tokensComplex State Systems (e.g., Inventory, RPG stats, Leveling)2,000–4,000 tokensComplete, Exportable Mobile Game (Multiple states, navigation, audio hooks)6,000–10,000+ tokens

Note: Token usage scales based on the complexity of your game's mechanics and the length of your active codebase.

🪙 Token Efficiency Tips

Maximize your subscription or free credits by vibe coding efficiently:

• Be direct: Use concise, instruction-heavy prompts instead of conversational filler.

• Batch your requests: Ask for a few related tweaks at once (e.g., "Change background to dark mode AND increase player speed by 20%") instead of making two separate runs.

• Clear the slate: If you are pivoting your game mechanics entirely, reset the AI context to prevent old, unused code from eating up your token window.

⚠️ AI Limits & "Hallucinations"

Vibe coding is incredibly fast, but AI models aren't perfect game engines. Keep these limitations in mind:

• Framework Drift: LLMs may occasionally use slightly outdated syntax or non-existent properties for newer mobile/React Native frameworks.

• AI Hallucinations: The AI might confidently write logic for an asset or API import that hasn't been defined yet, causing a runtime crash in your preview panel.

• Non-Determinism: If you feed the exact same prompt into the AI twice, it may build the underlying game state slightly differently each time.

🕹️ The Golden Rule: Always use the built-in Live Preview to playtest your mechanics, tap the buttons, and thoroughly break your game loops before exporting your code to Expo!