The Roleverse Engine · v4.2 "Ashfall" Core online · 99.98% uptime · 41 regions
The machine under the myth

The most advanced narrative engine on Earth.

Every world you fall into is rendered live by one system — a real-time stack that maps desire, directs pacing, voices a full cast and remembers every unsaid word. The same engine quietly powers writers' rooms, audio studios and streaming originals. Roleverse is just the most honest thing we built on top of it.

CORE v4.2
engine · live trace— region eu-fra-3
Stories spun · tonight
0
+1,240 in the last minute
Throughput
0M tok/s
sustained, all regions
Median latency
0ms
prompt → first scene
Live voices
0
neural · full-cast
Worlds online
0
and counting
(01) — Six subsystems, one heartbeat

What the engine is made of

Not one model — a coordinated stack, each layer specialised and obsessive about a single craft. They run together, in real time, for every scene.

SYS · 01

Desire Graph

A living taxonomy of 2,140 tropes, kinks and beats — mapped, weighted and traversable in real time. It knows that "snowed in" wants to sit next to "only one bed".

2,140 nodes94k edgeslive-weighted
SYS · 02

Scene Director

Models the heat curve scene by scene, so a slow burn actually burns slow. It decides what gets withheld, and for exactly how long.

per-scene pacingrestraint modeltension graph
SYS · 03

Voice Forge

Multi-cast neural voices with breath, restraint and timing. Every character sounds like a person who wants something — not a narrator reading.

38 voicesfull-castbreath-aware
SYS · 04

Continuity Core

Remembers every glance, scar and unsaid word across an entire arc. Nothing contradicts; a promise in scene one lands in scene five.

full-arc memorycross-scene0 contradictions
SYS · 05

Consent Lattice

On-device 18+ guardrails and per-reader boundary controls. Adults only, always, by design — limits are honoured before a word is rendered.

on-deviceper-readeraudited
SYS · 06

Render Loom

Generates and hand-grades cover art to match the exact mood of each world — then colour-matches it to the heat the Director set.

mood-matchedhand-graded3/4 covers
(02) — From a single line to a living world

One prompt, six passes, 240 milliseconds.

Input drops in. Every subsystem runs its pass. By the time you blink, a world exists — paced, voiced, continuous, consented.

Promptyour line
Desire Graph~28 ms
Scene Director~46 ms
Voice Forge~92 ms
Render Loom~58 ms
Your worldsealed
(03) — Fastest in class

Nobody else renders this fast.

Median cold-start latency, head-to-head. We clock a full world — desire-mapped, directed, voiced — in the time others take to return a single token.

Roleverse Engine0.24s
Atlas-72.10s
Mira LLM3.40s
Openverse4.80s
Legacy stack6.20s
(04) — Trusted by

The writers and studios who already know.

Velvet LabsThornfield PressNocturn AudioLatereader StudiosDuskmoth OriginalsOonling Works
"The pacing model alone is worth more than every tool I tried before it."
nocturne · 52 worlds · 1.2M readers
(05) — Under the hood

Full specification.

Versionv4.2 "Ashfall"
Parameters1.4T · sparse mixture
Subsystems6 · co-resident
Context memoryfull-arc, cross-scene
Voices38 neural · multi-cast
Languages24
Median latency<240 ms
Cold start0.24 s
Uptime · 90d99.98%
Regions41
Safetyon-device consent lattice
First shipped2021
The engine is running. Right now.

Your world is one tap away.