Seed a world — characters, a theme, a loose direction — and my-AI-story writes a coherent serialized story, episode after episode, narrated with a distinct voice per character. Every episode is verified against a series bible, so episode 7 still remembers episode 1.
The hard part of long-form generation isn't writing a page — it's writing page 200 without contradicting page 1. my-AI-story drafts a series bible (the canonical source of truth for characters, world facts, and the arc), then writes each episode against it and refuses to keep anything that breaks canon. Where my-AI-stro verifies knowledge, this verifies narrative continuity — same thesis, new medium.
The series bible — canonical characters, world facts, and the arc — is the source of truth. New canon (a death, a reveal) is folded back in through a verified update, so the constraints grow with the story.
A local LLM drafts prose and speaker-tagged dialogue for each episode; a local TTS engine renders a distinct voice per character and stitches them into one narrated episode.
Before a word is kept: continuity against the bible, structure of the beats, and speaker assignment — all blocking. No quiet contradictions; episode 7 still remembers episode 1.
my-AI-story holds a serialized world together across many episodes and renders it as multi-voice audio, entirely on local models — Ollama for synthesis, a local TTS for voices. Nothing leaves the machine.