AI

info

This is a preview API available for developers to test only. We'll enable it for production use soon.

Using generative AI in your games can lead to fun and surprising gameplay. Rune provides an easy-to-use API for submitting large language model (LLM) prompts based on user actions, with responses fed back into the game logic.

Using AI in your Game

To submit a prompt, call Rune.ai.promptRequest with a list of messages the AI should process. Under the hood, this is fed to the GPT-4o mini LLM model provided by OpenAI. The response is returned to the game logic through Rune.ai.promptResponse.

Simple Text Prompts

Example of a single plain-text prompt:

// logic.js
Rune.initLogic({
  actions: {
    myAction: (payload, { game }) => {
      Rune.ai.promptRequest({
        messages: [{ role: "user", content: "What is Rune.ai?" }],
      })
    },
  },
  ai: {
    promptResponse: ({ requestId, response }) => {
      console.log(response)
    },
  },
})

The input messages use the role key, which maps to the OpenAI API, so you can provide both user and system roles.

Conversational Prompts

Since the AI API is stateless, you pass multiple messages in a single request to maintain conversation context. You can store the prompts and responses in the Rune game state to maintain context like this:

// logic.js
Rune.initLogic({
  setup: () => ({
    messages: [],
  }),
  actions: {
    myAction: (payload, { game }) => {
      const message = { role: "user", content: payload.userQuestion }
      game.messages.push(message)
      Rune.ai.promptRequest({ messages: game.messages })
    },
  },
  ai: {
    promptResponse: ({ requestId, response }, { game }) => {
      console.log(response)
      game.messages.push({ role: "assistant", content: response })
    },
  },
})

Image-Based Prompts

The AI can analyze images provided as data URIs (external URLs are unsupported):

// logic.js
Rune.initLogic({
  actions: {
    myAction: (payload, { game }) => {
      const dataUri = /* Generate data URI, e.g., from canvas */
      Rune.ai.promptRequest({
        messages: [
          { role: "user", content: { type: "text", text: "What is Rune.ai?" } },
          { role: "user", content: { type: "image_data", image_url: dataUri } },
        ],
      })
    },
  },
  ai: {
    promptResponse: ({ requestId, response }) => {
      console.log(response)
    },
  },
})

What Makes a Good Prompt

AI prompt engineering requires experimentation and refinement. While there’s no single right way to structure prompts, here are a few tips:

1) Be Explicit

Define clear boundaries for AI responses to avoid unexpected behavior:

You are a cow. You may only respond with the word "Moo" or slight variations. Do not answer in human language or about topics a cow wouldn’t know.

2) Provide Examples

Include input-output examples and anti-examples to set expectations.

3) Use Ratings

Avoid yes/no questions. Instead, ask for weighted responses:

Rate the player’s interest in the conversation on a scale of 1 to 10.

4) Favor Unstructured Output

Avoid structured JSON responses for now. Instead, ask for specific formats and parse the result.

5) Animate During Prompts

AI responses take time—add animations or visual feedback to keep players engaged.

6) Account for Player Behavior

Players will test the limits of your AI. Anticipate challenges and ensure your prompts define acceptable behavior.

Rate Limiting and Costs

info

This is a preview API available for developers to test only. We'll enable it for production use soon.

The AI API is currently free for testing. Once in production, AI API usage will deduct Rune credits from the Creator Fund earnings.