I’ve been working on an AI-powered RPG where the story evolves in real-time based on player choices. The goal was to combine traditional RPG mechanics (HP, XP, inventory, progression) with dynamic LLM-generated narratives.

Some of the interesting challenges we ran into:

• Maintaining story consistency across long sessions

• Preventing narrative drift when users take unexpected actions

• Managing structured game state alongside free-form text

• Handling token limits for multi-turn storytelling

Approach:

• Using layered prompts (system + state injection)

• Summarizing past events to maintain continuity

• Separating narrative generation from game logic

• Tracking player state externally instead of relying on the model

Still experimenting with improving coherence over long sessions and reducing repetition.

Would be interested to hear how others are handling:

• Long-term memory in interactive apps

• Narrative consistency in branching flows

• Cost vs quality trade-offs for real-time generation

We also opened a small beta if anyone wants to test the experience:

Hi Zack! I hope you are well.

I am a journalism master’s student at Univiersity of the Arts London, and I am currently working on an article about AI RPGs and immersive experiences. I would love to get your perspective as a person who has made an AI-powered RPG (more specifically, on the technical side and what makes these experiences so compelling for users). If you’re open to it, I’d love to have a quick chat with you about your development process and where you think the whole space is headed.

Looking forward to hearing from you!

I am working on something similar for a while. Have GitHub pages describing the project (not yet published). I am unable to paste here project pages describing the project in detail (GitHub static pages)

So, the pipeline is getting quite a complexity using orchestrated main pipeline, RAG for lore, history where I use conversation semantic search, memories, etc. Tiered memory system with gradual compression, subagents to act on behalf of active NPCs, weather simulation, map support and detailed scene etc. tracking. Works reasonably well, but it is quite token demanding (using ollama cloud mostly) and lazy (1 turn about 60-90 seconds as of now)