inferred from GitHub
muni
Charactger AI App - Made for Anthropic Opus 4.7 Hackathon
Core trick
Persona-prompted model replies include an emotion label that selects which pixel-art sprite renders with each message.
Evidence
README matched ""Cerebral Valley" "Opus 4.7"": README.md
README excerpt
# Character Chat > Talk to historical figures. They don't break character — and they think harder when you do. Built for the **[Built with 4.7 Hackathon](https://cerebralvalley.ai/e/built-with-4-7-hackathon/)** (Cerebral Valley × Anthropic, April 2026). Pixel-art conversations with **Frida Kahlo**, **Sigmund Freud**, **Simone de Beauvoir** and **Salvador Dalí**, powered by **Claude Opus 4.7** with adaptive extended thinking. --- ## What it does Pick a character. Chat with them. They reply in-persona, in your language (EN/ES), with a pixel-art emote that matches their reaction (neutral / happy / thinking / surprised) and a tiny stage direction describing the scene. - **Frida** paints from pain in the Casa Azul. - **Freud** lights a cigar in his Vienna study and asks about your mother. - **Beauvoir** sips coffee in a Paris café and dismantles your assumptions about freedom. - **Dalí** dissolves the conversation into a melting Catalan beach. Ask any of them about Python. They will refuse — beautifully, in character. --- ## The Opus 4.7 angle: **Adaptive Extended Thinking** Most "chat with a character" demos either don't use extended thinking at all, or burn a giant thinking budget on every message regardless of context. Both feel wrong: - **No thinking** → philosophical questions get reflex answers that feel shallow. - **Always-on thinking** → "hola" makes the model spend 6,000 tokens deliberating before saying hi. We classify each user turn into a tier and scale Op