Why I Brought MyAdventures Back
Where it started
Back in late 2023, I set out to solve a problem that's plagued tabletop roleplaying fans for years: how do you get a genuinely good solo TTRPG experience?
The dream was to use AI as a real dungeon master, not just a chatbot that vaguely improvises, but something that could hold a world together. Interwoven lore, persistent adventure memories and precedent, unfolding story arcs, threads that actually mattered across sessions. The kind of continuity that makes a world feel alive.
MyAdventures went live in early 2024.
The 4,000-signup problem
Within a relatively short time, the site had attracted around 4,000 signups. People were using the free tier. Nobody wanted to pay.
And honestly? They had good reasons not to.
The experience I was chasing required feeding the AI enormous amounts of context, all that lore, all those memories, all those story threads. Token counts were climbing into the tens of thousands per request. In 2024, that felt excessive, and it genuinely was costly. But here's the rub: even at that cost, it still wasn't delivering the experience I was after.
The other half of the problem was tool calling. The kind of structured, reliable AI behaviour you need to run a proper interactive fiction system, querying memory, tracking state, and making consistent decisions about the world, was just not dependable enough outside the very top-tier models from OpenAI and Anthropic. Mid-tier models fumbled it. The economics didn't work. The experience didn't work. I took the site down in mid-2025.
What changed in the meantime
I didn't stop working on it.
The core codebase has been largely rewritten. When I run it against the top models now, Claude Opus, for instance, the experience is genuinely fantastic. Not "pretty good for an AI." Actually fantastic. The kind of solo TTRPG session I originally imagined is now fully achievable.
The problem, of course, was that Opus-tier inference is still expensive. A great experience you can't afford to run sustainably isn't a product; it's a proof of concept.
Why now
The landscape is shifting.
Models like GLM are bringing capable, cost-efficient inference into reach in a way that simply wasn't possible a year ago. For the first time, the gap between "this works brilliantly" and "this is viable to run" is closing. We're not fully there yet, but we're close enough that it's worth opening the doors again and testing properly.
So MyAdventures is back. Not as a finished product, but as a beta. I want to find out where the rough edges are now, what the experience feels like with the new architecture, and whether the economics finally make sense.
If you were one of the original 4000, welcome back. If you're new here, I hope it's worth the wait.
— David