Every hosted story service makes the same trade on your behalf: it keeps the model, the memory and your words on someone else's machine. That is not a bug in those products. It is their business model.
FPV starts from the opposite assumption — that the place where a story remembers you should be yours to choose.
The meter is the difference
Cloud narration is metered because it has to be: someone is paying for the token you just spent. Local narration is not. When the model runs on your computer, the only costs are the ones you already own — electricity, hardware, the time a model takes to think.
There is no credit counter in the corner of the screen, because there is nothing to count.
Memory lives next to the model
Long stories fail when the context that matters is far from the thing generating the next line. FPV keeps the story database, the Codex and the model on the same machine. The narrator can consult what happened twenty turns ago without the whole archive traveling over a network first.
That proximity is not a performance detail. It is what lets a scene respond to something that happened three sessions ago.
The escape hatch stays open
Local-first does not mean local-only. If you want a larger cloud model for a specific story, you can connect your own key and switch that story to Cloud allowed. The important part is that the default is the private path — and the choice is per story, visible and reversible.
What this means in practice
Run a smaller model for everyday scenes and a bigger one when a scene deserves it. Your stories never stop being yours. The trade is simple: a little setup once, in exchange for no account, no meter and no one else's server in the middle of your fiction.