Hidden Door updates its story engine with a tool-using AI Dungeon Master

A DM that both tells and prepares

Hidden Door says every experience on its platform runs on a narrative world model, which represents the state and narrative trajectories of all the people, places and things in a world. The update this week targets the engine that drives each turn of play.

The company frames its design around a tabletop Dungeon Master, a role it splits into two jobs. The first is telling the story in the moment: describing the scene, voicing characters and reacting to what players just did. The second is preparation, planning just enough ahead that, whatever the players do, there is good material to work with, such as a secret waiting to come out or a rival who holds a grudge.

According to Hidden Door, most AI storytelling only does the first job, while its harness is built to do both. The company argues that hand-authored branching cannot deliver real consequence, because a writer can only author so many branches and paths eventually fold back into the same few outcomes. It says it instead works with narrative setups, tropes, arcs and resolutions.

How the harness works

At the center is a strong, tool-using model. Rather than improvising everything from one giant prompt, it calls on tools when needed. Hidden Door lists four:

• World context: the people, places and rules of the world being played in.
• History: what has happened so far, including facts and hooks from earlier chapters.
• Quality: a check on how well the current scene is working as a story.
• Narrative playbooks: hand-written directions for telling a chunk of story well, such as revealing a secret, a slow burn seduction or a satisfying brawl.

Each turn, the DM pulls the context it needs from the world model, judges how well the scene's narrative is working and adjusts, and follows the relevant playbook. It then offers a menu of possibilities for the next moment while updating longer-term planning and game state. Finally, it receives the player's choices and intentions for their character and decides what should actually happen next.

Because the system is always planning a little ahead, it can hold a detail from an earlier chapter and bring it back at the right moment. Hidden Door calls this Chekov's Armory, where every open thread is tracked and used when it can best pay off.

Less structure, more small tools

The company says the new system has much less enforced structure than its previous ones, and more small, clear tools. In its view, this makes the harness more adaptable to the complexity of the narrative and to player choices.

The human-written playbooks are described as key. Without them, Hidden Door says, LLM stories tend to become samey. The playbooks are where its narrative designers impose their own approaches to storytelling on the system.

Hidden Door also draws a comparison to coding tools. Harnesses such as Claude Code and Codex give a model tools for planning, writing and reviewing code. Hidden Door says its version gives a model tools for context, quality, planning and human-written storytelling, with the same architecture aimed at a different goal: immersive stories.

What else is in the engine

The harness is one part of a larger story engine. Hidden Door names three other pieces of technology:

• Fast and slow architecture: generative games need to deliver results to players in seconds while also solving hard problems.
• Fractal memory: a system for storing narrative context by moment and by long-term trajectory.
• Combinatorial story tropes: the system plays into expectations for a trope, then subverts it while combining it with other tropes.

The company says it will explain each of these in the coming weeks. For now, its advice is that the best way to see the difference is to play. The post offers no benchmarks or user data, so the improvements described remain the company's own claims.