How we use AI at Storyarn

How we build Storyarn

We decide what to build and in what order. Before a larger change, the agent proposes a plan, and nothing is built until we approve it. AI coding agents write most of the code and its tests, and help write the documentation and translate it and the interface into Spanish. They follow rules written down in the repository on conventions, architecture, accessibility and security. Today they run on Claude models, by Anthropic, and GPT models, by OpenAI.

Every pull request runs through automated checks: thousands of tests, end-to-end tests in a real browser, type checks, a security analysis and the rules that keep the architecture in order. Our exports to Ink and Yarn Spinner are compiled with those tools' own compilers. Other AI agents review the changes too, and we read their reviews.

A change reaches the Storyarn you use only when we merge it, and we try new features in the product ourselves before we release them.

User data and development AI

We never use what our users create in Storyarn, such as their projects, Sheets, Flows, Scenes or workspaces, to develop it with AI. The AI agents we use never read that content or the production database. For development they use local data that belongs to Storyarn's creator, Adrián Nuhacet, and to diagnose problems they can read error reports, server logs and usage analytics.

How we make our demo

Afterself, the cyberpunk story in our screenshots, is the demo project we are building in Storyarn. We decide its world, its characters and every turn of its plot. AI agents help write the design documents from those decisions, write the dialogue and build the project: its Sheets, Flows and Scenes. We review each Flow in the editor. The characters' portraits and the maps are made with AI image generators, and we choose every image.

How we write our articles

We choose each article's topic and angle. An AI agent writes the first draft, in English and in Spanish, and checks every fact, quote and link against its sources. We read the whole text, review how each claim was checked, rewrite whatever is wrong or does not sound like us, and decide whether it is published. Later revisions follow the same steps.

Every article ends by saying how it was made: its sources and the history of its revisions, with the date of each one, the model that drafted it and the person who reviewed it. Screenshots of Storyarn are real captures of the product, and experiences and quotes come from real people and real sources.

Why we work this way

Storyarn is built by one person. It began as a side project, and because of how much it does and the approach it takes, we are now turning it into a commercial product: we believe it is worth it. For now it is free, and no investors fund it. AI agents let one person build and maintain a tool that brings together Sheets, Flows, Scenes, Brainstorming, localization, version control and real-time collaboration, and keep their own time for the decisions that shape it.

Speed is only worth something if the result is correct, so we care as much about checking as about writing: the rules the agents follow, the tests and reviews that changes go through, and the time we spend using the product ourselves.

Storyarn is made for people who write stories. We use AI for what it does well, such as writing code, drafting and checking, and the decisions stay with people: what we build, where a story goes and what we publish. We say so openly, so that anyone who uses Storyarn or reads what we publish can judge it knowing how it was made.

Who is responsible

Adrián Nuhacet is responsible for this policy, for how we build Storyarn and for everything we publish. Contact: hello@storyarn.com.

If you find a mistake

Email hello@storyarn.com. If an article needs fixing, we fix it and add the change to its revision history, with the date. If you see something that does not match this page, tell us too.