How Claude AI Powers the EchoQuest Game Master

When you type "I draw my sword and demand the merchant explain himself," something remarkable happens. Within seconds, a fully contextual narrative response appears — one that remembers who the merchant is, what happened three scenes ago, what your character's personality is like, and even the price you tried to haggle to in the previous market visit. The merchant might back down. He might call for the city guard. He might recognise you and ask, with terror in his voice, why you've come back. None of these responses were written in advance. So how does that work?

This post is a plain-language tour of what's actually happening under the hood when you take an action in EchoQuest. No deep machine-learning theory required. By the end you'll have an accurate mental model of what the AI Game Master can do, what it can't, and why EchoQuest is structured the way it is.

The Old Way: Decision Trees

Text adventures and early visual novels worked by mapping every possible player input to a predetermined response. If you typed "go north," the game checked a table and returned the "north room" text. If you typed anything else, you got "I don't understand that." Designers tried to anticipate every reasonable command, then added more whenever a beta tester surprised them. There were tricks to widen the parser — synonym tables, pattern matching, "guess the verb" mechanics — but the underlying limit was the same: every meaningful response had to be written by a human in advance.

This was fine for simple puzzles, but it breaks down immediately when you try to have a conversation, act creatively, or do anything the designer didn't anticipate. The world felt hollow because it literally was — it only contained what someone explicitly programmed. The classic frustration of "I can see the rope on the table but I can't pick it up" wasn't bad design; it was the parser running out of pre-written content. AI Dungeon, around 2019, was the first big public attempt to escape that ceiling using language models. It worked — but only loosely. State drifted, characters forgot their names, plot threads dissolved. The next leap, the one EchoQuest is built on, is what happens when you wrap a much more capable language model in a structured game system that holds the world together.

The New Way: Language Models

EchoQuest uses Claude, a large language model developed by Anthropic, as the core of the AI Game Master. Claude doesn't work from a lookup table. Instead, it understands your input as natural language and generates a contextually appropriate response from scratch, every time. The model has been trained on an enormous amount of text — fiction, dialogue, technical writing, philosophy, jokes, recipes, cultural references — so it has an unusually deep grasp of how human conversations and stories work.

That means:

• You can phrase your actions any way you want — "draw my sword," "unsheathe my blade," "ready my weapon," or even "let him know I'm not joking" all land in roughly the same place
• The GM understands intent, not just keywords — it cares about what you're trying to accomplish, not which exact verbs you used
• Responses feel natural and varied rather than canned — the same situation, played twice, will produce two different paragraphs of narration
• The story can go in directions nobody predetermined — the writer of the world only had to set up the conditions; the model improvises within them

The flip side: a language model on its own doesn't remember anything between calls and doesn't enforce rules. It will happily contradict itself if you let it. That's why the EchoQuest GM is more than just Claude — it's Claude wrapped in a game engine that supplies memory and constraints.

What the GM Actually Knows

Before Claude generates a response to your action, it receives a detailed system prompt containing:

• World information: the setting, tone, factions, and lore from your campaign's Game Bible. This is the same Game Bible the world creator wrote, condensed and structured so Claude can keep the world consistent without using up the entire context window on backstory
• Your character: name, class, stats, backstory, current inventory, conditions like "wounded" or "exhausted," and any relationships you've established with named NPCs
• Current context: where you are, what's around you, what time of day it is, what's happened in this scene so far
• Conversation history: the last several exchanges in your session, plus a summarised digest of older events that fall outside the live window
• Rules: how to handle combat, skill checks, NPC behavior, and narrative pacing — these are the EchoQuest "house rules" written into the GM persona

This is called the context window — everything the AI knows before it writes its next response. The richer and more relevant the context, the more coherent and immersive the story. A lot of EchoQuest's engineering work goes into deciding what to include. Ten thousand tokens of pristine, scene-relevant context will produce far better narration than a hundred thousand tokens of unfiltered transcript.

Why Responses Feel Consistent

One of the challenges with AI Game Masters is maintaining consistency — the same NPC shouldn't forget your name between scenes, the laws of the world shouldn't arbitrarily change, and a story flag set in chapter one should still be set in chapter five. EchoQuest addresses this by:

• Storing key story facts and character state in a structured game state separate from the conversation history. HP, inventory, faction reputations, completed quests, and named NPC dispositions live in a database — they're injected fresh every turn rather than depending on the model "remembering" them
• Injecting a summary of past events when sessions continue after a break. If you played three sessions last week and come back, the GM gets a bullet-point recap of what happened so it can pick up exactly where you left off
• Using the world's Game Bible as a persistent ground truth the AI always references, so the model can't quietly drift off-canon. If your world says elves are extinct, the GM won't introduce a friendly elven shopkeeper in scene fourteen
• Re-injecting the active scene's location description before any combat or skill check, so spatial details stay stable even when the model's attention has moved

The combined effect is a session where the GM feels like the same GM across hours of play. NPCs you met early in the campaign greet you by the name they used last time. The kingdom's politics evolve in directions consistent with what you've already established. The world has continuity even though, technically, every response is generated from scratch.

The Human Design Behind the AI

Claude doesn't invent the rules — we do. The EchoQuest team writes the system prompt that defines how the GM should behave: how to pace tension, when to offer choices versus let the player freeform, how harsh or forgiving to be with consequences, how to handle player goals that conflict with the campaign's central tension, when to say "yes, and" and when to push back. We've iterated on that prompt for hundreds of hours, comparing how different versions handle the same scenarios, and we keep refining it as players send us moments that didn't land the way we wanted.

Think of it like a very skilled, very fast co-author. We set the creative constraints. Claude fills in the story within them. The AI is genuinely creative — surprising even us with NPCs, twists, and lines of dialogue that we'd never have thought to write — but it's creating inside a frame we shaped. That's the secret to an AI GM that consistently produces the kind of story you signed up for, instead of drifting into generic fantasy mush by hour two.

What This Means for You

When you sit down to play, you don't have to think about any of this. You type or speak whatever your character does, and the world responds. But knowing what's underneath helps you play to its strengths. Be specific about what your character wants. Reference earlier moments — the GM is listening. Ask NPCs questions you genuinely don't know the answer to; the model is at its best when there's room to surprise you. And remember that the world has its own logic: if something feels off, it's worth asking the GM about it in-character. Often, what looks like an inconsistency is actually a thread waiting to be pulled.

Ready to see it in action? Play a free session →