Meet Nutmeg, a voice-first stage-manager mouse. She takes kids from “I’m bored” to a printed kit and a show, then steps back so the play happens off the screen.
See how it works ↓
Kids talk with Nutmeg. A grown-up only needs to press Print.
Pick a game and a theme just by talking. Nutmeg offers a few ideas, asks how many kids are playing, and hands out a role to each one.
Go on a prop hunt around the house: pillows, chairs, blankets, boxes. Only soft, safe items, and a missing prop can be swapped or skipped.
An AI-generated custom kit to color and cut: role badges, a sign, play money, tickets, a menu and prop cut-outs. Up to 7 pages, one print for a grown-up.
Nutmeg guides the show scene by scene with picture checklists and tablet hand-offs (“Pass the tablet to the Captain!”), then steps back.
Every kit is generated for the game the kids just planned: thick-outline line art they can color, dashed cut lines, and a blank “Name: ____” line on each badge that kids fill in themselves.
No accounts, no backend, no recordings, no transcripts. Nutmeg only asks how many kids are playing. Voice audio is processed live by OpenAI under its API terms, and PretendThat keeps none of it.
A native iPad app (iPadOS 17+) with no third-party runtime dependencies.
OpenAI Realtime API, speech-to-speech over a WebSocket with AVAudioEngine. Tool calls drive the app’s state machine.
Anthropic’s Claude writes each kit and the scene-by-scene play guide as structured JSON.
OpenAI image generation draws every kit page and the sticker art for ideas, roles and props.
Layered vector Nutmeg animated in SwiftUI, with a mouth that follows the voice’s volume.
A Swift package holding the app’s logic, covered by about 200 unit tests.