Story AI on Android

The two narrator choices on a phone-hosted world, a human Dungeon Master or your OpenAI key, how the key covers every campaign on the device, what it costs, and why local models and agent programs are desktop only.

Android app

The Story AI choice belongs to the phone: pick once and every campaign on your device world follows it, new or old. The Story AI button on the home screen reopens the screen, and inside a campaign the Setup tab shows the device's choice.

A human Dungeon Master

No AI. A person runs the table from the DM seat, on this phone or on another device that joined, with the console, the binder and the map tools. Everything else, dice, sheets, initiative, maps, voice, is the same.

AI Dungeon Master via your OpenAI API key

Create a key at platform.openai.com on an account with billing set up.
Paste it into the API key field. The Dungeon Master model (default gpt-5.1) and the cheaper utility model for summaries (default gpt-5-mini) are filled in; change them if you like.
Save. The key is stored by the server inside the app, never shown to players, and covers narration and pictures for every campaign on this phone. The field then reads "Saved on this device. Blank keeps it".

The phone does no model work: each turn, the server inside the app sends the prompt (the story so far, the sheets, the scene) to OpenAI and streams the narration back. Costs are OpenAI's per-token prices for the models you chose; a long session with pictures on is where the cost goes, and pictures can be turned off per campaign in the Setup tab. See OpenAI and compatible keys and What leaves your machine.

Why only two doors

A local model needs a desktop GPU and tens of gigabytes, and an agent program (Claude Code, Codex, opencode, Grok Build) is something a computer runs. Both doors exist on the desktop app. A phone that joins a desktop-hosted world or a self-hosted server gets whatever narrator that host has, so a group with one capable computer can still play from phones with a fully local AI.

Other backends

OpenRouter, a llama.cpp or Ollama server on another machine, ComfyUI at an address on your network, and narration (Kokoro, or any OpenAI-compatible speech server) are configured through the world's admin panel (account menu, Admin) on the phone, the same as on a self-hosted server: a text model section with a backend URL, model and key, and image and speech sections. The Story AI screen covers the common case; the admin panel covers the rest. A human Dungeon Master's dictation uses the phone's own speech recognizer.