FRONTIER AI
FOR LIVING WORLDS.
Built with the Mirari API
CHARACTERS / YOU CAN BUILD
Every persona below is one API call away — memory, voice, and a consistent personality, in any art style. Built-ins to remix, or author your own.
りんThe Spark
Rin
ツキThe Oracle
Tsuki
モチThe Host
Mochi
The Operator
Vex
The Sage
Sable
The Rogue
Kit
THE PLATFORM · MIRARI API
LIVING CHARACTERS & WORLDS, AS AN API.
One endpoint for characters that remember, speak, and stay in character — and the worlds they live in. Built for games, companion apps, and agents.
CHARACTERS
Personas with persistent memory and a consistent voice. They recall past sessions, hold opinions, and evolve with every interaction.
WORLDS
Multi-character scenes with shared state. Characters know each other, the setting, and what just happened — so the world stays coherent.
VOICE
Real-time, low-latency speech with cloned and original voices. Every character sounds like itself, in any language.
MEMORY
A managed memory layer per character and per player. No prompt-stuffing, no vector plumbing — it just remembers.
SEE IT LIVE
MEET THEM IN MIRARI.
Mirari is our own app, built entirely on the platform — AI characters with memory you can chat and talk with, plus community-made worlds. The best demo of what you can build.
GET MIRARI FOR iOS ↗RESEARCH
BUILT ON FRONTIER AI RESEARCH.
Forerunner publishes its work. Our research spans mixture-of-experts architectures, reinforcement-learning dynamics, and the long-context memory systems that keep characters coherent — tested on models with hundreds of billions of parameters.
WHY FORERUNNER
THE HARD PARTS, SOLVED ONCE.
A believable character is a systems problem — memory, voice, safety, and scale, all at once. We solve them at the model layer so you don't have to.
Persistent Memory
Characters remember across sessions without you managing context windows or vector stores. Recall that feels human, not retrieved.
Real-Time Voice
Sub-second, expressive speech with original and consented cloned voices — streamed, interruptible, and natural.
Safety & Consent
AI disclosure, content controls, and a Voice ID Registry so every voice is used with permission. Trust is the default, not an add-on.
Built to Scale
The same elastic GPU stack that serves our own app serves your traffic — from your first player to millions.
BY THE NUMBERS
0M
TOKENS OF CONTEXT / CHARACTER
<0
MS TO FIRST TOKEN
0+
VOICE LANGUAGES
0/7
ALWAYS-ON, PERSISTENT WORLDS
WHAT YOU CAN BUILD
ONE API. EVERY KIND OF CHARACTER.
01GAME NPCS
Drop characters into your game that remember the player, react to the world, and never repeat the same line twice.
02COMPANION APPS
Ship a voice-native companion with real memory in days — the hard parts (persona, memory, voice) are the API.
03AGENTS & WORLDS
Author whole casts and scenarios. Let players step into worlds that keep running whether or not anyone is watching.
