Charts and maps
Three kinds of node draw what flows into them. A Vega-Lite node draws a chart from a JSON specification, an Autark node draws maps and plots of city data, and a Simple View shows a table or a set of pictures with nothing to write.
Vega-Lite charts
Connect a table, such as a pandas DataFrame or a GeoPandas GeoDataFrame, to a Vega-Lite node and write its specification in the Grammar tab. Curio supplies the rows, so the specification leaves out data and names columns directly:
{
"mark": "bar",
"encoding": {
"x": { "field": "neighborhood", "type": "nominal" },
"y": { "field": "vegetation_pct", "type": "quantitative" }
}
}Press play to draw it. If the editor is still empty when data first arrives, Curio fills it with a starter specification chosen from the column types. For a GeoDataFrame, "mark": "geoshape" draws a map, and Curio picks the geometry column and a projection unless you set them yourself. A chart with several views keeps its size, and the node scrolls to show all of it. Once a chart has a specification, it redraws by itself whenever new data arrives.
Autark maps
An Autark node takes one JSON specification, called an UrbanSpec: data loads tables, compute runs passes over them, and map and plot draw them, as maps of layers in 2D or 3D, colored by a column, and as plots such as histograms. A node with no map or plot passes its tables on by their names (such as table_osm_buildings), for another node to draw or read. A table from another node arrives as a layer named after the input circle it comes in on, input_0 for the first, here colored by its mean column:
{ "map": { "layerRefs": [{ "dataRef": "input_0", "getFnv": "mean", "getFnvType": "quantitative" }] } }A raster, such as a GeoTIFF from the Data Catalog or the output of a Raster Calculator, draws as a raster layer colored by one of its bands, band_1, band_2 and so on: { "dataRef": "input_0", "getFnv": "band_1" }.
A layer's title names it in the legend, and color draws it in one color. A line layer takes a width in the map's planar units (about three quarters of a meter at Chicago), a polygon layer a stroke color, a points layer a pointSize, the radius of each point in meters, and a raster layer colorMapReverse and colorMapCenter to reverse its color scheme or center it on a value. A layer colored by a number runs its colors from its lowest value to its highest, or over colorMapRange, [low, high]: maps that set the same range color a value alike. A layer whose color is its input's scenario chip, such as "[!! scenario input_1 !!]", is drawn in the color of the scenario box that input comes from (see Maps and differences).
layerType draws a layer's table as another type of its kind of geometry, such as lines as roads; "legend": true puts a layer's colors in the map's legend; and pickedComps and skippedComps draw features as picked as the map opens, or hide them, by their positions. The Autark node section of the usage guide, linked below, covers these and more.
Maps, plots and compute need a browser with WebGPU, such as a recent Chrome or Edge. Without it, the node says WebGPU is not available, and its Check again button asks the browser again.
Several views
A node draws every map and plot its specification writes, each in a pane of its own: the maps first, then the plots. layout arranges the panes: "vertical" stacks them, "horizontal" puts them side by side, and "grid" fills columns columns (2 by default), row by row, with gap pixels between panes. An id names a map or a plot, and links joins a plot to a layer of a map: a brush on the plot lights the same features on the map, and a pick on the map selects them in the plot.
{
"layout": { "type": "horizontal", "gap": 8 },
"map": { "id": "buildings", "layerRefs": [{ "dataRef": "table_osm_buildings", "isPick": true, "getFnv": "height" }] },
"plot": { "id": "heights", "dataRef": "table_osm_buildings", "mark": "bar", "axis": ["height", "@transform"], "transform": { "preset": "binning-1d" }, "events": ["brushX"] },
"links": [{ "source": "heights", "target": "buildings" }]
}The node keeps its size, and its views share it. Each map has its own menu and legend, and R frames again the map you clicked last. In a node of several views, each view with an id also has an interaction dot of its own, which links that view alone to other nodes (see Which views take part).
Data and compute
The sources of data run in order, and each makes or changes a table that later sources, passes, maps and plots name:
osmloads OpenStreetMap layers, such asbuildingsandroads, from an extract (pbfFileUrl) or from OpenStreetMap's Overpass API, where itstagSetsalso pick features by their tags, such as cafés.csv,jsonandgeojsonload a file, or rows written in the specification, andgeotiffloads a GeoTIFF as a raster.heatmapbins a table into a grid, with a band for each entry of itsgroupBy, such as a count of the table's rows withinnear.distanceof each cell's center or the mean of one of their columns. With nogroupBy, its one band counts the rows. A heatmap is a raster.joinadds to each feature of a table the rows of another that meet it or lie withinnear.distance, aggregated by itsgroupBy;sqlsaves the rows of aSELECTover the tables before it as a new table;updatechanges a table's rows; anddropremoves a table.
A heatmap, a join, an update or a drop also reads a table of the node's input by its name, such as "joinDataRef": "input_0" for the points a Python node returns. Autark cuts every table to the area of the first table in its workspace, so load another area, such as a second city, with a workspace of its own: a map, a plot or a pass reads the tables of the workspace it names.
Each feature of an OpenStreetMap layer has its OpenStreetMap id in osm_id, and a building's parts share its building_id. Tags that hold numbers, such as height, building:levels, maxspeed and lanes, are numbers: lengths in meters, speeds in km/h and counts as plain numbers. A value written with a unit, such as 282 ft or 30 mph, is converted, and any other value, such as none or 4 floors, is left empty, as in a Discovery Catalog download. A building part with no height stands 6 m tall above its base wherever the node uses a height: the map draws it so and a compute pass reads it so, while a plot shows the heights its table holds. "buildingsZeroHeight": true, on a buildings layer, draws such parts at random heights of 3 to 7 floors instead, and passes read no height for them.
Each entry of compute is a pass, run in order. A GPGPU pass runs a function written in WGSL on the GPU, once for every feature of a table. A render pass draws the tables its layers name, also on the GPU, from each feature of its viewpoints table, and counts what the frames show, such as the share of sky a street sees between the buildings. On the CPU, a windows pass places the windows of a buildings table, and a clusters pass numbers the groups of rows whose shapes touch. A pass writes into its table, or into a new one named by outputTableName, which later passes, maps and plots read by that name. With "writeBack": true, a pass also keeps the table it writes in the node's database, where a source with "after": "compute", run once every pass has run, reads it: a join, a heatmap or a SQL query of a pass's result.
Camera, terrain and style
A map opens framed on what its layers draw, looking straight down on their middle. Its camera turns that view: pitch tilts it, bearing turns it to face a compass heading, zoom sets how far away the eye is, center is the [longitude, latitude] it looks at, and motion is a tour the map plays once it opens, which a drag, a scroll or a key stops. Pressing R frames the map again, turns it by its camera and plays its tour again. terrain raises the ground to the heights of a raster's band, with the flat layers draped over it and the buildings standing on it. style names one of autk-map's styles, such as light, or gives the nine colors of a style of your own; highlightColor colors the picked features, and "showUi": false hides the map's menu and legend.
Plots
A plot draws a table's rows as its mark says, such as bars, a scatter or lines, over the columns axis names, and transform bins or sorts them first, as a histogram does. color colors the marks by a column, tickFormats and axisLabels set the ticks and the axes' titles, selection draws rows as selected, and events lists the selections a click or a brush sends. A plot also draws a table with no geometry, such as the rows of a sql source. markColor and highlightColor color a plot's marks and its selected ones: every Autark plot on a page draws in the last ones set, so give every plot of a dataflow the same ones, or none.
Simple View
A Simple View node needs no specification. It shows a DataFrame or GeoDataFrame as a table, or as one card per row when a column holds pictures, either links to images or images stored in the data; with several such columns, Image column picks which one to show. Anything else, such as a number, a list or JSON, appears as text with a Copy text button.

In the example gallery
The examples in your projects show each of these at work. "Vega-Lite GeoDataFrame maps" draws maps with Vega-Lite, "Autark PBF loading" and "Autark GPU shader" build Autark maps, "Simple View: tables and images" covers Simple View, and "Heterogeneous data + linked views" puts an Autark map beside Vega-Lite charts that respond to each other (see Linked views). "Autark: Back Bay or the Loop, whose streets get more sun at the solstice" loads two cities into two workspaces and draws four linked views in one Autark node. "Autark: park access heatmap" and "Autark: sidewalk problems in the Loop" bin parks and Project Sidewalk's labels into heatmaps, "Sky seen from the street" runs a render pass, and "Autark: green share and summer heat in Niterói" and "Autark: park area within reach and summer heat in Niterói" load a summer temperature GeoTIFF beside OpenStreetMap layers. The example gallery linked below has a step-by-step walkthrough for each one.

