Feature

Spaces

A space is the editable SIE environment for a real or simulated place.

Spaces hold the semantic layer that tells agents what exists in an environment: rooms, zones, objects, landmarks, routes, and other spatial facts. The agent does not need the raw scene file in its prompt. It needs grounded context from the space, such as which room is active, what object was recognized, what is nearby, or what route applies.

Spaces can start in a few ways:

from a video capture that SIE turns into an initial space
from an imported WorldAgents semantic layer JSON file
from scratch in the space editor

After a space is created, teams can open it in the SIE space editor to rename it, inspect the environment, and edit the semantic layer. Those edits make the space more useful to agents because they turn visual scene data into named, reviewable context.

Spaces are still WIP, but the product direction is clear: a team should be able to capture an environment, review and improve its semantic layer, then let agents use that space for recognition, localization, simulation, and place-aware behavior.