Life Blind Box Runtime

Life Blind Box Runtime is a voice-first AI microdrama engine for short, replayable interactive story sessions. Instead of watching a fixed story, users speak, choose, and live through a five-round dramatic experience that ends with a personalized ending card. The current live demo proves the core product loop end to end: voice input, story generation, TTS narration, player choice, replay, and ending card completion. The public GitHub repository is a public-safe showcase edition with provider interfaces, public-safe scene templates, documentation, screenshots, and setup notes. Private production prompts, full scene libraries, secrets, and internal QA logs are intentionally excluded. The product thesis is that AI voice entertainment is easy to demo once, but hard to make replayable, affordable, and commercially scalable. Voice-first sessions create repeated ASR, LLM, and TTS calls. If every turn uses the most expensive model path, consumer entertainment margins may fail. Life Blind Box Runtime is designed for cost-aware model routing. Simple transitions can use cheaper routes, ordinary story branches can use standard model tiers, and emotionally important turns or ending-card moments can use stronger models. The live Railway demo proves the voice-first microdrama loop. Fireworks / AMD-compatible routing and cost telemetry are the next milestone, not a completed claim. This submission is for the Unicorn Track because it focuses on product completeness, startup vision, and the economics of an AI-native entertainment runtime.