Categories
Self-Hosted Local LLM Gaming Roleplay Tabletop RPG AI Dungeon Master Interactive Fiction WorldbuildingStars
Forks
Watchers
Developer links
Narrative Engine
Narrative Engine is an AI dungeon master that remembers every scene of a tabletop campaign, so session 200 can still quote the betrayal from chapter 3. Every turn is archived verbatim into sealed chapters, a two-step search recalls the exact scene, and auto-condensation keeps the context window under its limit. A world-state register tracks who is where, who holds what, and which secrets each NPC may know. NPCs get generated profiles, a six-axis personality, goals that advance on background dice rolls, and portraits in five art styles. World Arcs escalate storylines on their own, event engines add surprises and encounters, and a timeskip simulates the gap. Steering tools include swipe, continue, one-shot event injection, an out-of-character Ask GM chat, a Lore Check, and scene rollback that rewinds the whole world state. A character wizard, staged inventory changes, a captioned image gallery, and a location ledger with an unfinished pixel-art map round out the table. Extensions add features such as an enemy compendium, and example worlds ship ready to play. An encrypted vault protects your API keys, and the web interface opens from a tablet or phone on your network. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.
Benefits
- Memory that survives hundreds of sessions
- Every turn is archived verbatim and recalled by a two-step search over sealed chapters, so the GM can quote an exact line from a scene played two hundred sessions ago.
- NPCs that pursue their own goals
- Each NPC has a six-axis personality and goals that advance on background dice rolls, so characters scheme, drift, and collide while the player is busy elsewhere.
- Control over cost and context
- A live token gauge, three condensation strategies, an AI tier switch, and per-module token caps let a cheap model handle background work while the main model only narrates the turn.
- Your campaigns stay on your machine
- Campaign data lives in local files, semantic search runs locally, and API keys sit in an encrypted vault, so nothing goes to third-party vector providers or a cloud.
Features
- Lossless scene archive
- Every dice roll, line of dialogue, and narrative beat is stored verbatim in sealed chapters with generated overviews, then retrieved scene by scene using local vector embeddings.
- Divergence Register facts
- After each turn the engine extracts facts such as locations, promises, debts, and alliances, groups them by chapter and category, and records who knows each one.
- Living NPC ledger
- NPCs are detected as they appear and profiled with personality, voice, goals, faction, and looks, then tracked with relations, pressure meters, signature kits, and portraits.
- World arcs and event engines
- Background arcs advance rung by rung on dice, while surprise, encounter, and world event engines raise their odds every quiet turn and fire configurable events.
- GM steering toolkit
- Swipe or continue a reply, inject a one-shot event, ask the GM out of character, run a Lore Check on any message, and roll any scene back.
- Extensions and provider roles
- A frozen mod API adds features through the Extensions tab, and up to five OpenAI-compatible endpoints split story, summarizer, utility, image, and auxiliary work.