When a D&D campaign goes long enough (past session 30, past the second arc, past the point where everyone is on first-name terms with the NPCs), a particular kind of failure starts to happen. Not the dramatic kind. The quiet kind.
The DM forgets which thread they planted in session 12. The players forget what they did with the magic item from session 22. Someone says “wait, didn’t we already meet this person?” and nobody is sure. The shared fiction starts to fray, not from anyone’s neglect, but from the simple fact that nobody’s brain is built to hold eighty hours of narrative state.
I run two long campaigns. I have this problem. So do my players.
This is the part of running TTRPGs that AI is genuinely good at, and almost no current tool focuses on.
What I’ve already built
A few pieces of my Obsidian setup are AI-native skills that solve specific slices of running a campaign:
- A session prep skill that reads my campaign overview, surfaces what happened recently, and helps me build the next session’s beats
- An NPC generator that creates contradiction-driven characters with wants, secrets, and a voice, and saves them to the right campaign folder
- A decision logger that captures the reasoning behind worldbuilding choices so future-me knows why
These work. I use them. They’ve changed how I prep.
But notice what they don’t do.
The gap
None of them watch what happens during play. None of them track the slow accumulation of plot threads, items, and NPCs across sessions. None of them tell me, two months into an arc:
“You introduced a poisoner in session 14 and the party hasn’t followed up. The dwarf priest in session 19 hasn’t been mentioned since. There’s an unresolved promise to the Captain from the third session of this arc.”
That work is supposed to be done by my brain. And after a year of running, my brain is not doing it well. Neither are my players’ brains. The cognitive load of campaign continuity is the actual hard problem in long-running TTRPGs, and most existing tooling solves around it rather than for it.
World Anvil is excellent at worldbuilding: you fill in the wiki and it indexes it. It is not a continuity engine. Notion is flexible but offers no model of campaign state. The specialized DM tools focus on encounters, characters, and maps. The space between “everything written down” and “knowing what’s currently relevant” is mostly empty.
What it might look like
The narrow version of this:
- Ingests session transcripts (Granola, Otter, manual notes; capture is increasingly solved by other tools)
- Extracts campaign state: NPCs and their last appearance, items and their current holder, plot threads and their status, promises made and pending
- Maintains a queryable timeline that anyone, DM or player, can ask
- For the DM: surfaces what to bring back next session, who’s overdue, what’s dangling
- For the players: answers “what do I know about X” and “what happened in the last few sessions”
Not generation. Not advice. Just memory. “I’m not asking the AI what to do. I’m asking it what happened.”
That’s a much smaller, sharper product than “TTRPG copilot” or “campaign manager.” It’s also the part LLMs are genuinely good at right now: extracting structured relationships from messy narrative text, then surfacing them on demand. The hard problem isn’t the AI. It’s the product around it.
What I’d need to validate
I name the problem with confidence because I have it. The harder question is whether I’ve over-projected: whether other DMs running long-running homebrew campaigns at the level where this would matter actually share the pain, or whether they’ve quietly built their own workarounds the way I built my prep skill.
Three things I’d want to know before building anything:
- Frequency vs. tooling tolerance. How often does the continuity gap actually bite, and how rarely does it bite badly enough that someone would change tools?
- The privacy question. Sending session transcripts to a third party is the price of entry. Some DMs will balk. What’s the percentage who won’t?
- The integration story. Can this be a layer on top of where DMs already work (Obsidian, Notion, Roll20, Foundry), or does it have to be its own thing? The “yet another tool” tax is real and TTRPG audiences are unforgiving of it.
Status: hypothesis
I haven’t built it. I might. Right now this is the version of the idea I’d be willing to test in conversation with three or four DMs running long campaigns.
If you’re one of those DMs and any of this resonates (or doesn’t), I’d want to hear from you. The thing I’ll commit to either way is publishing the gap honestly before claiming the solution.
That’s already more than most “AI for X” pitches manage.