You’re three hours deep into a session. The paranoia is thick. You’re looking over your shoulder, double-checking every shadow, and piecing together a conspiracy that actually feels grounded—no simulation theory nonsense, no “it was all a dream” tropes, just pure, high-stakes tension. Your AI Game Master is weaving a masterpiece of narrative tension. You type out your next desperate move, hit enter, and…
Your GM just walked out of the room mid-combat.
If you play solo TTRPGs using Artificial Intelligence, you know this wall well. In a previous guide, I broke down the theory of prompting an AI to be a ruthless, logically consistent Game Master. But theory only gets you so far when you prefer multi-session campaigns. The longer the game goes, the more context the AI has to juggle. Eventually, you hit the context window ceiling, or worse, you burn through your daily and weekly message limits at breakneck speed.
I used to run my entire multi-session campaigns in Claude. Its prose is exceptional, and it understands narrative pacing better than almost any other model. But I was hitting my limits fast. I had to improvise.
What emerged from that frustration wasn’t just a workaround; it was a completely new, infinitely scalable workflow. It’s a multi-LLM pipeline that uses different tools for what they do best, ensuring the lore stays rock-solid, the conspiracy remains grounded, and the game never has to pause for a server timeout.
Here is the exact three-phase workflow I use to run infinite, deeply connected solo campaigns.
The Core Problem: Memory vs. Prose
To understand the pipeline, you have to understand why a single AI struggles with long-term solo play.
When you’re running a game built on conspiracy and paranoia, continuity is everything. The AI needs to remember the breadcrumbs. If an NPC lied to you in Session 1, that lie needs to pay off in Session 8. But Large Language Models are stateless. They don’t “remember” anything; they just re-read the entire chat log every time you send a new message.
If you use a model with a massive context window to hold all your sessions, it gets expensive, slow, and eventually, it loses track of the fine details. If you use a standard model, you hit rate limits because every message you send is forcing the AI to process tens of thousands of words of backlog.
You cannot rely on the engine generating the prose to also act as the long-term archive. You have to separate the brain from the memory.
Phase 1: The Spark (Claude)
Every campaign needs a strong foundation. Even though I don’t use Claude to run the entire campaign anymore, it remains my absolute favorite tool for inception.
Claude is brilliant at establishing atmosphere. When you are kicking off a new game—setting the tone, defining the parameters of the mystery, and getting your character into their first real slice of trouble—you want top-tier prose.
The Workflow
Claude has done its job. It has generated a rich, highly detailed, text-heavy foundation for the world. Now, we have to save it.
Phase 2: The Vault (Google Docs)
This is the simplest step in the pipeline, but it is the structural load-bearing pillar of the entire operation.
You take the raw, messy transcript from your Claude session—every dice roll, every piece of dialogue, every out-of-character clarification—and you copy it directly into a Google Document.
Why this matters
You are creating an immutable “Source of Truth.” AI models are prone to hallucination. If you leave the lore up to the AI’s internal reasoning, a gritty street-level conspiracy will eventually mutate into a cosmic matrix narrative. By exporting the session to a Google Doc, you are locking the narrative in stone.
I usually title these documents clearly:
Clean up any glaring typos if you want, but don’t worry about making it pretty. This document isn’t for human eyes; it’s food for the engine.
Phase 3: The Continuity Engine (NotebookLM)
This is where the magic happens, and how we completely bypass the rate limits of standard chatbots.
Google’s NotebookLM is not a standard conversational AI. It is a RAG (Retrieval-Augmented Generation) system. You upload documents into a “Notebook,” and when you interact with it, it explicitly and exclusively reads from those provided sources.
This makes NotebookLM the perfect surrogate Game Master for long-running, grounded campaigns.
The Migration
Eventually, you want to turn these raw, mechanically heavy transcripts into something publishable for a blog—something that reads like a novel rather than a math textbook.
This is where you bring the raw Google Docs back to a standard conversational AI (like Claude or Gemini) for post-processing. Because you are no longer using this AI to run the game, you aren’t fighting rate limits or context windows. You are simply giving it a specialized, one-off task.
The Bridge Prompt:
Take the following TTRPG session transcript and rewrite it as a cohesive, third-person omniscient narrative chapter. Maintain the grounded, paranoid tone. Do not skip any major story beats or environmental details. Remove all visible mechanics, dice rolls, and out-of-character chatter. Translate mechanical successes/failures into descriptive prose. [PASTE TRANSCRIPT]
Why NotebookLM Excels at TRPG
When playing games where you constantly have to look over your shoulder, you need an impartial referee that doesn’t forget your past actions. NotebookLM’s strict adherence to the source documents is a superpower here.
Because it actively queries your uploaded text, the continuity is staggeringly precise. It grounds the narrative entirely in what has already been established.
For example, when I am playing, I don’t have to waste token space reminding the AI of my exact stats or recent history. If I declare an action, NotebookLM inherently knows my character’s background, stats, and inventory. It remembers that a rival gang brings up past grievances from a specific event because it’s right there in the text.
It even perfectly tracks mechanical nuances. If I take a hit, I don’t have to correct the math—it knows I’m sitting at exactly eight stress, not nine, by cross-referencing the character sheet and the session log.
It anchors the game. The conspiracy doesn’t spiral into the stratosphere because NotebookLM cannot invent world-breaking lore that contradicts the Google Docs you provided.
The Infinite Loop (Playing and Updating)
Once you shift gameplay over to NotebookLM, you have bypassed the traditional rate-limit wall. You can play for hours. But to keep the campaign alive, you must maintain the Vault.
Your AI Game Master just successfully “slept” and committed the entire session to its permanent, unchangeable long-term memory.
When you sit down to play Session 3, NotebookLM will read both Session 1 and Session 2. It will know who you killed, what bullets you spent, and which locked doors you left behind.
The Final Step: The Weaver
Eventually, you want to turn these raw, mechanically heavy transcripts into something publishable for a blog—something that reads like a novel rather than a math textbook.
This is where you bring the raw Google Docs back to a standard conversational AI (like Claude or Gemini) for post-processing. Because you are no longer using this AI to run the game, you aren’t fighting rate limits or context windows. You are simply giving it a specialized, one-off task.
The Translation Prompt:
By separating the Engine (the AI running the game) from the Weaver (the AI writing the prose), you optimize both. The Engine stays fast, mechanically strict, and flawlessly consistent via NotebookLM. The Weaver gets to flex its creative muscles, turning a successful ‘tinker’ check on a rusted train door into a tense paragraph about sparking wires and scraped knuckles.
Closing the Loop
Running solo campaigns shouldn’t feel like you are fighting the technology. You shouldn’t have to hold the entire plot in your head just in case the AI forgets.
By chaining these tools together—using Claude for the initial atmospheric spark, Google Docs as the immutable vault of truth, and NotebookLM as the memory-perfect Game Master—you create an environment where you can finally just be a player.
You can focus on the conspiracy. You can look over your shoulder. And most importantly, you can keep playing, knowing your Game Master isn’t going anywhere.
