Case study
TARN
Stories born at the table.
TΛRN is an AI memory and story layer for tabletop RPGs. The table plays the system it already plays; TΛRN records the session, keeps track of the campaign and turns what happened into comic-style chapters that the game master reviews and publishes.
The challenge
The best stories never leave the table.
Every session of a tabletop campaign produces story, and most of it survives only as patchy notes and memories. Recaps take effort, and nobody outside the group ever gets to read what happened.
The idea
Don’t build another RPG. TΛRN has no rules, no dice and no world of its own: it is a memory and story layer over the system a group already plays. The table plays as usual; TΛRN listens, remembers and turns the session into a chapter.
What we built
Recording with consent
TΛRN records the session in the background while the table plays. Recording needs consent from everyone at the table, and every session asks the table first.
Campaign memory
It keeps track of who was there, what they did and what it meant, across sessions rather than one at a time.
Game master review
TΛRN proposes the important moments. The game master includes or ignores each one and confirms what becomes canon.
Chapters, not transcripts
Pick Chronicle, Cinematic or Epic, choose a summary or a full chapter, and TΛRN writes and illustrates it as comic pages with lettered speech balloons.
Publishing on your terms
Keep a chapter private, share it with the table or publish it for readers. Nothing leaves without the game master’s say.
A reader for everyone
Readers discover stories in a For You and a Following feed and read one full-screen page at a time, free and with no account needed.
How we built it
TΛRN is a single TypeScript monorepo. A Next.js app serves the website and the API for both the browser and the iOS and Android app, which is built with Expo. Shared packages hold the API contracts, the domain rules (canon, permissions, limits) and the design tokens, so web and mobile run on the same definitions.
Turning a recording into a chapter is a long job, so it runs as durable background workflows: understanding the session, finding important moments, building the story, preparing scenes and checking continuity. A step that fails retries on its own instead of starting the session over, and players can close the app while it works.
- AI in the product
- AI models turn the recording into text, work out what happened and draft the script and the illustrations. Every model output is validated before it is saved, and a person stays in charge: the game master decides what becomes canon and what gets published. Lettering is set in code over the artwork, not painted into it, so balloons fit their text.
- AI in the process
- Built with AI coding agents working from a versioned specification and architecture decision records kept in the repository. Changes are checked with type checks, linting and tests.
Stack
Status & what’s next
TΛRN is open as a beta at tarn-steel.vercel.app. Anyone can read published stories, and game masters can create an account and start a campaign with no card. The service is still in development, and the iOS and Android apps are in testing, not yet in the app stores.
What’s next
- Launch on its own domain, tarn.quest.
- Bring the iOS and Android apps out of testing and into the stores.
- Keep testing with game masters at real tables and refine how chapters read.
