Solo RPG campaigns fail at session four. Mine always did, and after two years of asking around, I am pretty sure most people's do too. The chat tool forgets your first mate. The notebook is technically accurate but emotionally inert, or becoming increasingly hard to maintain. The summary you wrote last week was good enough then and almost useless now, because the parts of it that mattered are not the parts you summarized.

I have spenta long time assembling and tearing down a personal stack for solo play. Ironsworn here, Mythic over there, a chat window I treated like a sparring partner, a notebook that always ran out of pages at the wrong moment. Every time I sat back down, I had to rebuild context for 10-20 minutes before I felt adjusted enough to make a decision that mattered.

With that the campaigns drifted and the characters thinned out. I would lose the thread of who someone was, the genesis or spark of their creation, and the next session would be the wrong story about the wrong person. That experience is the whole reason Wanderhold exists. What follows are five frustrations that became design decisions, in roughly the order I had to face them.

1. Context collapse is what marks a solo campaign short

I've seen it called "Context collapse". It's the phrase people on the solo subs use when they describe what happens after about session four-five. The chat tool forgets your first mate. The summary is accurate and inert.

My first fix was longer summaries. Then structured character sheets pasted into every prompt. Then a "previously on" template I made myself read at the top of each session. Each one worked but the longer the game, the more there are and the longer the notes get.

They all failed at the same moment, which is when a decision from week two should have weight in week six and instead just vanishes into the recap. The thing I actually wanted was a system that holds the campaign for me. Not a chat history or log, with hallucinating AI. A structure where the second time my character meets a particular crew member, the system knows what happened the first time, what was said, what was risked, who paid, and what is now different.

Wanderhold's first design constraint was: state lives in the engine, not in the prose. If you sit down a month later, the engine knows your character. The story does not have to start over. There's a whole construct and memory that the engine runs and stores, so that the narrative always makes sense.

2. Being your own GM is a tax I stopped wanting to pay

There is a phrase that gets quoted a lot: taking off the GM hat. Anyone who has run a solo session knows what that means. You are playing your character, running the world that opposes your character, managing the rules between them, and deciding whether the scene is interesting yet.

That is four jobs. They each use different parts of the brain. Doing all four at the same time, for two hours straight, is the part of solo play that gets hard fast.

Early on an oracle engine or written procedure can tell you which hat to wear when. I thought GM overhead was a discipline problem but it was not. It was a cognitive load problem, and no procedure I wrote was going to make it cheaper to keep swapping hats every thirty seconds.

The split I wanted is the one a real table gives you. The GM holds the world. The player makes choices. The rules sit between them and arbitrate. As a player, you do not have to invent the next complication or judge whether the dice you just rolled should matter.

Wanderhold is, more than anything else, an attempt to give a solo player that division of labor. The engine holds the world and proposes the scene. The mechanics decide what happens when you act. You stay in the seat of the player.

3. Free-form chat and structured RPG are different products

This one took me longer to admit than it should have. For a long time I was convinced I could get there with a smart enough prompt and a strict enough roleplay contract with the model. It just doesn't work. Every session I ran that way ended the same way: I rolled a one, the story made it okay, and I felt nothing.

I wanted to like a good chat-based story session. Writing into a model that pushes back is one of the better solo creative tools we have right now (and I am past pretending otherwise). But it is not the same thing as running a campaign.

The difference comes down to whether a failed roll actually changes what gets generated next. In a chat tool, the dice are theater. You can roll a one and the story will accommodate you, because the model is trying to be useful and the useful thing in a story is usually for the protagonist to land the swing.

That is fine if the goal is co-writing. It is corrosive if the goal is playing a game where the stakes are real and your decisions actually impact the story.

I wanted the kind of stakes that come from a mechanically enforced miss. The kind where you wanted the door to open quietly and it did not, and now the rest of the scene reorganizes around that fact, and the system, not you, decides what reorganized looks like.

That is a different architecture. It is a game with an engine that simulates pressure on you, not a writing partner that releases pressure to be agreeable. Wanderhold is in the second category. I respect the first category. They are different products, and the confusion between them is one of the more expensive and fast moving things happening in the space.

4. The oracle is the soul of the system, not a d100 table

When I started, I assumed the oracle layer of a solo RPG was basically a random table. Roll a number, get a result, write the scene. That is how most oracle systems present themselves on the surface anyawys, so that is what I tried first.

It does not produce texture though. It produces noise. My first oracle was a flat weighted table feeding straight into a prose prompt, and the results read exactly like that: an arbitrary thing happened, and then the model dressed it up.

What I ended up building is closer to a full narrative pipeline. A mission type rolls. An objective rolls inside that type, scoped to your world and your character's history. A story archetype creates scene beats that each also have oracle look-ups. A complication rolls against the active tone. A thematic axis is selected. Only then does the prose layer get the assembly and write the scene.

The randomness happens in stages, with each stage constrained by what came before, so the result is specific instead of arbitrary. Two campaigns running the same mission type produce different scenes because the inputs to the prose stage are different at every layer above it. I will write up the full staged-oracle architecture in a follow-up post - it's complicated to represent but I have some good visuals and interactive diagrams already done.

A weak oracle gives you a strong model writing prose around a thin spine, and you can feel it inside three sessions. A strong oracle gives the prose model something real to work with. That is the difference between a story you remember and one you closed the tab on.

5. The world has to be specific or none of it works

The temptation, when you are building any kind of generative narrative system, is to make the worlds broad. Cast a wide net. Generic fantasy. Generic space. Generic horror. The thinking is that breadth equals replayability, and specificity will narrow and/or isolate the audience.

I think that is exactly backwards, and I had to play test two generic worlds before I believed it.

The campaigns I remember from my own actual play are the ones where the world had teeth. Where the period was named, or the politics were specific. Like where the technology or the systems had a history. A generic frontier is a backdrop.

A frontier that is specifically post-war, specifically haunted by the particular legacies of a particular decade, with specific factions running specific operations across specific terrain, is a place where a decision can mean something.

Wanderhold launches with four distinct worlds, and each one is opinionated. They sit in particular literary traditions. They have ideological weights. The character types you can play in one world would feel wrong in another, because each world has been built to reward different kinds of choices.

I wanted players to feel, the first time they start a campaign in one of these worlds, that someone had decided what this place is. Not that the system is willing to be anything. That it is specifically this thing, and it is going to put pressure on you in this particular way.

---

Those five reasons are also five design constraints, and most of the engineering decisions in Wanderhold trace back to one or more of them. The engine holds state because of reason one. The mechanics fire automatically because of reason two. Dice rolls have teeth because of reason three. The oracle is staged because of reason four. The worlds are opinionated because of reason five.

I built the thing I wanted to play. I do not think it is the only good shape for an AI-driven solo RPG. I do think it is the right shape for the players who have been quietly assembling DIY stacks for years and would like to stop.

If you want to see whether any of the above lands in practice, you can start a free campaign without a credit card.