State snapshot

getState() returns a summary of the current scene as a single object. The AI console puts this whole thing into the system prompt, so the model can judge with knowledge of "what's loaded right now, where the view is, and what's selected".

oniyanma.getState()

Since it's read-only, it's usable as-is even on the read-only surface.

Fields

Display

Field	Type	Content
backend	string	Render backend (webgl / webgpu)
colorMode	string	Color mode (rgb / intensity / elevation / classification)
navMode	string	Nav mode (orbit / walk)

Layers

Field	Type	Content
sources	{ id, name, url?, visible, pointCount }[]	Loaded point cloud sources
surfaces	{ id, name, visible }[]	Reference surfaces (TIN)

id is what you pass to removeSource / setSourceVisible / setSurfaceVisible.

Selection and edits

Field	Type	Content
hasSelection	boolean	Whether there's a selection
selectionCount	number	Selected point count (resident-based)
editCount	number	Non-destructive edit count
editsByActor	{ human, ai, api, unknown }	Count per actor. unknown comes from an old save that didn't record an actor
canUndo / canRedo	boolean	Whether your own edits can be undone / redone

Note that canUndo is per user. It's false when there's a state with only other people's edits while collaborating. → Non-destructive editing

Coordinates

Field	Type	Content
alignedBounds	[minX,minY,minZ,maxX,maxY,maxZ] | null	The overall AABB in the aligned frame. null when nothing's loaded
cameraReal	[x, y, z]	Camera position (real coordinates, m)
frame	{ offset: [x,y,z], alignYaw } | null	The coordinate frame. scene = Rz(-alignYaw)·(real - offset). null when nothing's loaded
crs	{ epsg?, name?, verticalDatum } | null	The coordinate reference system. null if it couldn't be determined

verticalDatum is 'orthometric' (elevation) / 'ellipsoidal' (ellipsoidal height) / 'unknown'. CRS is recorded only — never reprojected. → Coordinate systems

alignedBounds is in the aligned frame; cameraReal is in real coordinates. Don't mix them.

Findings

Field	Type	Content
findings	{ types, total, published }	Number of finding types / total findings (including drafts) / count visible to the client

Counts only — no content. Since the snapshot rides in the system prompt on every request, this is kept from ballooning as findings accumulate. Get the content with listFindings.

On a build with no findings store, this key is omitted entirely (readable as "the findings feature isn't present"). → Filing findings

Example

{
  "backend": "webgpu",
  "colorMode": "elevation",
  "navMode": "orbit",
  "sources": [
    { "id": 1, "name": "kasado.copc.laz", "url": "https://data.oniyanma.jp/kasado.copc.laz",
      "visible": true, "pointCount": 95169313 }
  ],
  "surfaces": [],
  "hasSelection": true,
  "selectionCount": 128340,
  "editCount": 12,
  "editsByActor": { "human": 9, "ai": 3, "api": 0, "unknown": 0 },
  "canUndo": true,
  "canRedo": false,
  "alignedBounds": [-142.3, -88.1, -12.4, 142.3, 88.1, 31.7],
  "cameraReal": [-52341.2, 18904.7, 42.1],
  "frame": { "offset": [-52400.0, 18850.0, 20.0], "alignYaw": 0 },
  "crs": { "epsg": 6669, "verticalDatum": "orthometric" },
  "findings": { "types": 1, "total": 4, "published": 2 }
}

Querying individually

getState() is a summary — use query* when you need more detail.

What you want	Command
Resident point count and extent per layer	queryLayers
The selection's AABB	querySelection
Elevation distribution	queryElevation
Classification histogram	queryClasses
CRS and coordinate frame	queryCrs