My goal is to build a tabletop RPG (TT RPG) system that’s purely text-based, procedural, and decentralized, emphasizing reflection over modeling, exploiting LLM-native behaviors for illusionary world-building, and ensuring narrative cohesion without a centralized simulation. The “world” is an illusion in each player’s mind, generated from personal threads and minimal threading with proximate characters—think subjective narratives where consistency emerges from text reflections, not shared state.
First versions are alpha via https://onen.ai/
Core Design Philosophy:
-
Reflect, Don’t Model: Use LLMs to procedurally generate narratives as reflections of player prompts, evaluations, and echoes from others. No physics, economies, or global rules—everything is illusionary text.
-
LLM-Native Exploitation: Soft LLMs for mechanics eval, medium for personalized narratives/echoes, hard for background continuity/threading (e.g., updating subscriptions based on reflections).
-
Narrative Cohesion, Not World Building: Cohesion is minimal and emergent: Proximate engagements (small groups) via direct threading; far/observable via broadcasts. Background hard-LLM ensures continuity across narratives without enforcing a “true” world.
-
Character-Based Architecture: Each character is a self-contained “node” (accessible via browser), handling its own storage, processing (LLMs), and interactions. Redundancy via replication; nodes can run different versions, updated manually by players when prompted.
-
Effortless Scalability: Starts with handfuls of players, grows organically. No bottlenecks—use client/edge compute, P2P-like links for proximate, broadcasts for far. Simulate decentralization on cloud for persistence.
Use Case: A TT RPG where players submit prompts, LLMs generate cohesive narratives. Characters subscribe to others for initiative windows (group actions) or narrative feeds (echoes/reflections). System includes prompts, evals, dice (local), narratives, and hard-LLM for subscription updates/continuity.
I want to explore how to adapt Solid (Pods as character nodes) for this:
- Pods as Character Nodes: Each player’s Pod stores threads, summaries, attributes; runs LLM logic (commerical or locally run LLM).
- Browser Access: Players interact via web app, authenticating with WebID.
- Subscriptions/Broadcasts: Use Solid’s linked data for proximate engagements (access grants for threads), notifications/WebSockets for real-time, broadcasts for far illusions.
- Scalability/Redundancy: Pods self-operational on cloud hosts (e.g., Inrupt); replicate data to proximate Pods; handle version drift with protocol compatibility.
- Integration with LLMs: Embed LLM calls in Pod logic (e.g., via JS in Solid apps) for procedural gen; background hard-LLM as scheduled Pod tasks.
Straight up, I am old, can’t code, but I am starting to use Claude and Lovable to create working apps. I’ve been aware of SOLID, and only because I am trying to create a character-based architecture, that I remembered SOLID and I might embark on trying to set up the game through SOLID. The onboarding for a new player on onen.ai is relatively painfree, with optional engagement through Discord currently. How easy will it be for a player to sign up and start playing via SOLID? Some basic questions: Does a SOLID app need to be created, or can it run via pods directly? Please correct me if this is completely the wrong direction!