Built-environment records
Architectural plans, landscape layouts, housing documentation, mechanical sheets, infrastructure drawings, and related project material where rights and source permissions allow.
Data sourcing
We help teams get the strange, real data that is not already on Hugging Face.
ReplayAI focuses on source-driven datasets: messy real-world documentation, spatial material, video, and interaction traces that are useful for models learning to understand environments and act inside them.
Spatial and world-model data
Real spaces, plans, constraints, layout variants, and visual context for systems that need to reason about rooms, paths, buildings, interfaces, and physical structure.
Video and agent traces
Ego-view recordings, screen or interface interaction data, task traces, and custom collection plans for vision, robotics, and computer-use agents.
Processing and preparation
Filtering, redaction, conversion, organization, annotation planning, and GPU-scale preparation for teams turning raw material into training-ready datasets.
Rights, source permissions, and redaction are reviewed first.
eric@replayai.org