WORLDS FROM
NOTHING.
FRONTIER AI LAB
WE BUILD THE MODELS THAT BUILD YOUR WORLDS.
Forerunner.ai is an applied AI research lab building the full stack of generative spatial intelligence. We combine on-device vision-language models for instant scene understanding, serverless GPU reconstruction across multiple parallel pipelines, and 3D Gaussian splatting to turn ordinary photos into photorealistic worlds you can walk through.
Point your camera at any space. Our pipeline captures geometry, depth, and lighting, reconstructs a 3D Gaussian splat in seconds on serverless GPUs, and delivers it optimized for iPhone, Apple Vision Pro, and the web. One photo, every platform.
GENERATION PIPELINE
FROM MEMORY TO SPATIAL WORLD
CAPTURE & INPUT
AR-guided orbital capture with on-device vision AI. Real-time pose tracking, automatic keyframing, and semantic segmentation — all running locally on CoreML. No cloud round-trip for scene understanding.
RECONSTRUCT
Frames hit serverless GPUs where multiple reconstruction paths run in parallel — single-image splat synthesis in under a second, multi-view 3D with segmentation, and dense monocular SLAM for room-scale environments.
DELIVER
Metal-based Gaussian splat rendering on iPhone, RealityKit immersive spaces on Vision Pro, WebGL for the browser. One pipeline output, optimized and compressed for every platform.
RESEARCH
OPEN RESEARCH AT PRODUCTION SCALE
We publish our work. Our research spans mixture-of-experts architectures, reinforcement learning dynamics, neural 3D reconstruction, and spatial AI. Tested on models with hundreds of billions of parameters.
PLATFORM & TECHNOLOGY
EVERY WORLD RENDERED.
Our platform orchestrates on-device vision AI, serverless GPU reconstruction, and real-time 3D rendering across a multi-path pipeline — so you can go from camera to walkable world in seconds.
Gaussian Splatting
Photorealistic 3D reconstruction from photos. Our pipeline generates, trains, and renders 3D Gaussian splats optimized for real-time mobile exploration.
On-Device Vision AI
A bundled vision-language model runs entirely on-device for instant scene understanding, object detection, and semantic segmentation — zero cloud latency.
Serverless GPU Inference
Heavy reconstruction runs on serverless GPUs. Multiple paths execute in parallel and auto-select the best reconstruction.
Spatial Delivery
Metal rendering on iPhone, RealityKit on Vision Pro, WebGL on the web. One pipeline output compressed and optimized for every platform.
Your memories, reconstructed. Every world generated and delivered in seconds.
THE PLATFORM
0
OUTPUT PLATFORMS
<0s
PHOTO TO 3D WORLD
0
STEP GENERATION PIPELINE
0%+
3D COMPRESSION
CAPABILITIES
THREE PILLARS OF SPATIAL CREATION
CAPTURE
AR-guided orbital capture with automatic keyframing. On-device AI understands the scene before a single byte leaves your phone.
RECONSTRUCT
Serverless GPUs run parallel reconstruction paths — single-image synthesis, multi-view 3D, and dense SLAM. The best result is selected automatically.
EXPERIENCE
Walk through your worlds on Vision Pro, explore on iPhone, or share a link. Photorealistic Gaussian splats rendered natively on every platform.
