Every idea, woven into a story you can hear, see and shape.
A crew of AI agents writes it, voices every character, paints each scene and scores it with music — live, while you listen. Choose what happens next, or just say it.
Every idea, woven into a story you can hear, see and shape.
A crew of AI agents writes it, voices every character, paints each scene and scores it with music — live, while you listen. Choose what happens next, or just say it.
The crew is gathering.
Open a book to read it again, page by page. Stories are kept in this browser only.
StoryWeaver is a prototype of an immersive storytelling agent. Below: what we mean by immersive, the agents that make it happen, how a story flows through them, and the design decisions — and trade-offs — behind it.
Immersion is the moment you stop noticing the app and start living the story. Five things get you there:
Ten roles, each with one job. Only two call a language model for each story; plain code and your browser do everything that should be instant and predictable. The curtain rises on them one by one — tap anyone in the line-up to bring them back into the spotlight.
One LangGraph workflow on a stateless Python API: the Guardian and the Storyteller work in parallel, the Editor checks every line, and the whole story — both endings included — reaches your browser before it begins. While you listen, nothing waits.
Children may be listening, so safety is designed in — not bolted on. Deterministic rules come first because they are instant and can't be argued with; models handle nuance.
A prototype is a set of choices. These are the ones that shaped StoryWeaver.
From the evaluation suite (safety red-team, a single-prompt baseline comparison, reading level and timing). Details in the README.