xb-ai - SKILL.md Agent Skill

name: xb-ai description: >- Integrate generative AI into an XR Blocks app via `xb.ai` — text and multimodal (image) queries to Gemini or OpenAI, real-time Gemini Live audio/video sessions, and image generation. Use to answer questions, describe what the camera sees, drive conversational agents, or generate content from an XR scene. Covers `enableAI()`, `xb.ai.isAvailable()`, `xb.ai.query()`, `startLiveSession()`, `generate()`, and local API-key handling (`?key=` / `keys.json`). Always guard calls with `isAvailable()`; never ship keys in production client code.

xb-ai: Gemini / OpenAI integration

xb.ai (a.k.a. xb.core.ai) wraps Gemini (default) and OpenAI. See templates/6_ai, templates/7_ai_live, and demos/xrpoet.

Setup & keys

const options = new xb.Options();
options.enableAI();
xb.init(options);

For local prototyping, supply a key via the ?key=YOUR_KEY URL parameter or a keys.json ({"gemini": {"apiKey": "…"}}) served next to the app.

[!IMPORTANT] The ?key=/keys.json paths are for prototyping ONLY. In production, proxy AI calls through a server you control — never embed a key in client code.

Text & multimodal queries

class Ask extends xb.Script {
  async ask() {
    if (!xb.ai.isAvailable()) return; // always guard
    const res = await xb.ai.query({prompt: 'Write a haiku about dust.'});
    console.log(res.text); // response text

    // Multimodal: text + image parts
    const res2 = await xb.ai.query({
      type: 'multiPart',
      parts: [
        {inlineData: {data: base64Png, mimeType: 'image/png'}},
        {text: 'What do you see?'},
      ],
    });
    console.log(res2.text);
  }
}

Gemini Live (real-time audio/video)

await xb.ai.setLiveCallbacks({
  /* onmessage, onopen, … */
});
const session = await xb.ai.startLiveSession(/* LiveConnectConfig */ {});
xb.ai.sendRealtimeInput({
  /* audio/video chunk */
});
// xb.ai.getLiveSessionStatus(); xb.ai.stopLiveSession();

Capture the scene to feed the model with xb.core.screenshotSynthesizer / xb.core.deviceCamera (passthrough; enable via options.enableCamera()).

Image generation

const result = await xb.ai.generate('a low-poly desert fox', 'image');

Notes

Choose the model via options.ai (model: 'gemini' | 'openai', per-model options in src/ai/AIOptions.ts).
xb.ai.isLiveAvailable() gates Live; not all models/keys support it.
AI data leaves the device for Gemini/OpenAI servers — follow their privacy terms.