Spaces and maps

A space is the workspace you open and share. A map is one projection of one dataset inside it. A space can hold several maps, and every panel in the workspace is pointed at exactly one of them at a time. Floor is the old name for a map; it no longer appears in the interface, and if you came looking for a floor switcher there is not one.

Space

A space is what a URL like /space/<id> opens, and what a link, a visibility setting and a collaboration room all attach to. A space can be reached by its UUID or by a short alias.

The space owns:

• one or more maps,
• the space’s public or private visibility,
• the AI summary and metadata shown on the home dashboard,
• named state threads (bags, selection, pages, plot settings) synced through the Threads control in the menubar.

Map

A map is one dataset laid out in two coordinates. It has its own points, its own cluster tree, its own field types (which columns are titles, semantic text, numeric, categorical, dates) and its own statistics.

Each open map also owns its own scoped state in the browser: selection, bags, tools, view settings and point colors are per map, not per space. That is why selecting points on one map does not change the selection on another.

When a space is opened, the first map in its list becomes the active view and every panel follows it.

Switching maps

There is no floor switcher and no global map picker. Map selection is per panel, in the tab strip of the tabset:

1. Pick a map from the dropdown in the tab strip. Choosing a map locks that panel to it.
2. Use the target button on a map panel to point every unlocked panel at that map at once.
3. Use the lock button to pin a panel to its current map, or unpin it so it follows the active view again.

A locked panel keeps showing its map even after you focus a different one. This is how you put two maps of the same space side by side and give each one its own Inspector.

Adding a map to an existing space

Open View > Machine Learning > Space Creation and pick the Add Map tab. It lists the current map’s numeric metadata columns, with a search box and a regex toggle; select at least one, and Mantis builds a new map in the same space from those columns. You are redirected to the space with the new map’s synthesis progress showing.

This path derives the new map from data already in the space. It does not take a CSV upload. To bring in new source data, create a map from the create-space flow instead.

Floors

“Floor” was the original name for what the interface now calls a map. Two traces of it are still real, and both are below the UI:

• floor_number is a field on points and clusters in the API.
• connection is a column type the create-space API still accepts. Setting connection: true on a column marks it as the link back to the previous floor, and it is documented for API callers in the in-app developer docs.

Neither is exposed in the app. The column-type grid in the create-space flow offers title, semantic, numeric, categoric, date, links, image, geospatial, custom model, coordinate1, coordinate2, vector and delete. There is no connection checkbox, no floor selector, and no floor-to-floor navigation control anywhere in the interface. If a tutorial or an older page tells you to add a Connection column in the UI, it is out of date.

Pages

Separate from maps, a map can carry pages: alternative coordinate views built from two of its own metadata columns rather than from the embedding. Create them in Map settings; each opens as its own tab, and a page dropdown appears in the tab strip next to the map dropdown once a map has any.

Heads up

Pages render on the legacy map renderer. The default Map panel always plots the semantic embedding coordinates, so locking it to a page has no visible effect. Open the page from Map settings, which creates a tab that renders it correctly.

Read nextEmbeddings and projectionHow a dataset becomes the coordinates a map is drawn from.