TL;DR: AI as Your Guide, Not Your Game Designer
- MultiBit now has an AI agent built in. Ask it how the platform works, how to set up a component, or just ask it to make the change for you.
- This is not "type a prompt, get a game." The AI turns your ideas into configured components that run on a hand-coded game engine.
- Every change comes back as a proposal you review and approve before anything is touched.
- New components get placeholder art generated automatically, letting you play test fast, polish later.
- Want more horsepower? Connect Claude, ChatGPT, Grok, or any AI that supports MCP, and design and build your game with it directly.
Let's be honest. Nobody wants to read documentation.
You've got a game idea, you're excited, and the last thing you want is to spend an afternoon learning what a "workflow spec" is before you can make a lumber mill turn wood into boards.
In my last post, I talked about the new Edit Mode, which made building a game in MultiBit way easier. But even with a friendlier builder, there's still a learning curve. There are concepts to learn, and a bunch of ways to set things up.
So I asked myself: what if you could just ask?
Ask It Anything, Or Just Ask It To Do It
MultiBit now has an AI agent built right in. You can use it three ways:
- Learn the platform: "How do locations work?" "What's the difference between a skill and an item?"
- Get help configuring things: "How would I set up a bakery that turns flour into bread?"
- Have it do the work: "Add a bakery to this scene that turns flour into bread, and costs 50 gold to build."
That last one is the fun part. You can be in Edit Mode, looking at your game scene, ask for a new component, and once it's created, it just pops into existence right there in your scene. No forms, no hunting through menus.
This Is Not "Prompt a Game"
I want to be really clear about this one, because "AI" and "game builder" in the same sentence usually means something different.
MultiBit does not ask an AI to write your game. The game engine is hand coded. I started building it about two and a half years ago, and it's the thing that actually runs your game: the rules, the resources, the timing, all of it.
What the AI does is translate your ideas into configured components for that engine. It's more like a very patient assistant who has read all the docs and knows which buttons to press. Your game still runs on solid, predictable, tested code, not on whatever an AI happened to generate that day.
You're still the designer. The AI just speaks the platform's language so you don't have to learn it all up front.
How It Works (The Light Version)
For the curious, here's roughly what happens when you ask the agent for something:
- Figure out what you want. A small, very lightweight model reads your request and recognizes your intent. Are you asking a question? Creating something? Editing something that already exists?
- Look up what matters. Based on that intent, it pulls the relevant bits out of our knowledge base, the docs on how the system works, plus details about your game, and builds up the context for the next step.
- Plan it out. A planning layer takes that context and your request and creates a structured proposal: exactly which components to create or change, and how.
- You review it. Nothing happens until you look over the proposal and approve it.
- Make it real. Once approved, the proposal gets translated into the same server calls the builder uses to create and edit game components.
The neat part is steps 1 and 2. Because the agent only gets handed the docs that actually matter for your request, it doesn't need a giant, expensive model to do a good job. The whole thing runs on OpenAI's super lightweight gpt-4o-mini, and it does a remarkably decent job. Good context often beats a big brain :)
And that review step matters. The AI proposes, you decide. It's your game.
Auto-Generated Placeholder Art
I waffled on this one for a quite a while. Should the platform generate images automatically?
In the end, the answer was yes, because the alternative is a scene full of blank boxes or meaningless symbols. And nothing slows down realizing your game like having to stop and go find art for every new thing you add. When you're prototyping, you want to see your idea, not a grey square labeled "Bakery."
Again, I went with the lightest weight model that could do the job: gpt-image-1-mini. It can generate images with transparent backgrounds, so objects in your scene actually look like objects and not a bunch of janky rectangles. And the costs is incredibly cheap,about a penny per image.
Now, don't expect any masterpieces, this is placeholder art after all. The goal is to get you to a playable prototype fast, so you can playtest it and find out if it's fun. Polish comes later, once you know the game is worth polishing.
Design Cards: Your Game's Lexicon
Here's the catch with a lightweight model: it has a fairly limited context. It's great at "add this thing" and "how does that work," but it's not the partner you want for a long, winding brainstorming session.
That's where design cards come in (the one bit of scope creep I let myself have). Design cards let you write down your game pillars, key mechanics, story beats, and more. Over time, they become a kind of lexicon for your game: what things are called, how they fit together, and what you're going for.
They're a great way to plan out the structure of your game and design its mechanics. And since the agent can pull from them just like it pulls from the docs, and your actual game configuration, it gets a better sense of your game without needing to remember a two hour conversation.
Bring Your Own AI with MCP
But wait, there's more!
MultiBit now has an MCP layer. MCP (Model Context Protocol) is a standard way for AI assistants to connect to outside tools. In plain English: you can plug MultiBit into the AI you already use.
Here's how it works:
- Create a game in MultiBit.
- Grab your game's MCP credentials.
- Plug them into Claude, ChatGPT, Grok, or any AI agent that supports MCP connections.
Now you can learn about the platform, brainstorm, design, and even build your game with all the power of the big foundational models like Sonnet, Opus, Sol, and friends. Long brainstorming sessions, complex systems, big batches of components, the works. Those models can go as deep as you want, and they build on the same hand-coded engine as everything else.
Use the built-in agent for quick stuff while you're in the scene. Bring in the big guns when you want to go deep.
Want to Build One Yourself?
I'm building this platform so that you can make simulation, story and resource management games without touching code. And now you don't even have to read the docs first. Just ask.
If you're a creator, coach, or storyteller looking for a more powerful way to connect with your audience, or you just want to build a game without writing code, head to multibit.games to sign up for early access.
Your ideas, your design, your game.
The AI just helps you get there faster!