Date: March 9, 2026
Series: Dev Blog (Dev-05)
Project: Shattered Bounds: Battlegrounds (Ottawa, Canada)
TLDR
- Shipped: I built an Open Brain for the Shattered Bounds project over the last two weeks, bringing scattered planning, design, and lore into one retained system.
- Shipped: The Open Brain is already helping me keep project knowledge organized independently of any single LLM.
- Improved: That same structure now suggests a strong direction for AI NPC architecture, built around layered memory instead of one giant prompt.
- Blocked by: The NPC version is still a design architecture, not yet a fully implemented runtime system.
- Next: Define the first practical NPC memory stack and test it on a small example character.
What shipped / works now
The Shattered Bounds Open Brain exists
Over the last two weeks, I built an Open Brain for the Shattered Bounds project. The main value is simple: instead of leaving key ideas spread across chat threads, notes, and half-finished documents, I now have a more centralized way to retain and retrieve project knowledge.
That matters because Shattered Bounds is already too large to hold comfortably in short-term memory alone. It spans game design, business planning, worldbuilding, operations, hardware, and long-term technical architecture. Having that information organized in one place makes it much easier to stay focused and continue building without losing context every time I switch tools or revisit a topic.
The project knowledge is less tied to one AI model
One of the things I like most about this approach is that it reduces lock-in to any single LLM. If the knowledge is structured and retained outside the model, then the project itself becomes more durable. The information stays with Shattered Bounds, instead of living only inside whichever conversation happened to generate it.
For a long-term project like this, that feels important. I don’t want the memory of the world, the systems, or the decisions to disappear just because I change tools.
Reference: AI News & Strategy Daily / Nate B Jones
I want to thank Nate B Jones for putting out a video on Open Brain. This is what had inspired the open brain development for Shattered Bounds and my next step: the Shattered Bounds AI Brain. He has produced many useful and inspiring videos as well as a great set of articles on his substack. I am not affiliated with him or his channel but I do want to say thank you!
What improved
It changed how I’m thinking about NPC AI
While listening again to Nate B Jones talking about the Open Brain idea, something clicked for me: this may also be the right direction for NPC intelligence.
Up until now, I had not been able to wrap my head around a reasonable way to build NCP with consistent knowledge, memory and personality. You could write a big character prompt, include their backstory, maybe add some lore, and hope the outputs stay consistent. But that approach gets expensive fast, wastes context space, and doesn’t scale well once NPCs need to know about regions, factions, players, local events, and their own history.
The Open Brain approach suggests something better: give the NPC multiple memory layers, and retrieve only what matters at the moment.
The idea: layered NPC memory
The rough shape I’m imagining right now looks like this:
Static Reference Layers
These are the NPC’s pre-loaded knowledge layers.
This would include things like world lore, race-specific context, regional knowledge, and job-related expertise. Instead of hard-coding all of that directly into every prompt, the system could vectorize a kind of World Encyclopedia. Then, when a player asks about a local legend, a nearby conflict, or the customs of a certain people, the NPC retrieves only the relevant pieces.
That means the character can sound informed without dragging the entire encyclopedia into every interaction.
The Personality Layer
This is where things get more interesting.
The personality layer is not just flavor text. It acts more like a filter or retrieval bias. A cheerful NPC and a grumpy NPC may both have access to the same memory bank, but they do not surface the same things in the same way.
A cheerful character might be more likely to retrieve success stories, hopeful interpretations, or fond memories. A grumpy character might prioritize insults, failures, unresolved tensions, or reasons to distrust someone. The knowledge base may be shared, but the personality changes what gets pulled forward and how it is framed.
That feels much closer to how a character should work.
Background and Current Memory
This would be the dynamic read-write layer.
Every meaningful interaction could be embedded and stored as it happens. If a player helps an NPC in one town, that event becomes part of the NPC’s memory. If the two meet again later in another region, the system can semantically search that player history and retrieve the earlier interaction.
In practical terms, that means the NPC can remember the alliance, the favor, the argument, or the betrayal without me needing to manually hard-wire every possible branch. The memory becomes persistent, searchable, and portable.
For Shattered Bounds, that is especially exciting because the long-term vision is a living world. If NPCs can remember what players did, even in a limited but believable way, the world starts feeling less like a theme park animatronic setup and more like an actual place.
Why this matters
This feels important for two reasons.
First, it solves a development problem. Large prompts are clumsy. They burn tokens, become hard to maintain, and make consistency difficult. A layered retrieval model is cleaner and more scalable.
Second, it solves an immersion problem. Static NPCs are one of the fastest ways to make a world feel fake. If every character repeats the same information the same way forever, then no matter how beautiful the visuals are, the illusion breaks.
Shattered Bounds has always aimed toward a world that feels alive. I’m still a long way from the final version of that, but this Open Brain concept feels like one of the first architectural ideas that could actually support it.
What this means for players
For players, none of this is flashy yet. There is no big visual reveal here, no new monster, and no new arena mechanic.
But if this direction works, it could eventually lead to NPCs that feel much more believable. Characters would be able to speak from local context, respond in a way that matches their personality, and remember previous interactions instead of resetting to zero every time you meet them.
That would make the world feel more personal, and hopefully more alive.
Next Steps
- One sample NPC has a documented four-layer memory structure
- The retrieval flow is clear enough that it could be prototyped
- Static lore, personality bias, and player-memory examples are separated cleanly
- Implement a conversational NPC on this website for players to interact with
- Test build a Smallville test zone for NPC interaction
FAQ
Q: Is this already running in-game?
A: Not yet. Right now this is a design and architecture direction inspired by the Open Brain system I just built for the project itself.
Q: Why not just use one big character prompt?
A: Because that gets expensive, messy, and hard to scale. A layered retrieval model should be easier to maintain and better at surfacing only the information that matters in the moment.
Q: Does this mean NPCs will remember players?
A: That’s the goal. The dynamic memory layer is specifically aimed at allowing NPCs to retain meaningful interaction history and retrieve it later when relevant.
Q: How does this help Shattered Bounds overall?
A: It supports the long-term vision of a world that feels persistent and reactive, instead of one where every NPC behaves like a reset button.
Q: When can we see NPCs implemented?
A: A full implementation of NPCs is expected in the Shattered Bounds: Grand Dungeon expansion. Before that, you can expect to see NPCs on this website to interact with for story, as well as Shattered Bounds: Tower will contain some NPCs.
Help us build it
Donate: http://ko-fi.com/liminalrealms
Partner: contact@shatteredbounds.com
Leave a Reply