Node graph

Every action in the 3D viewport becomes a node. The graph lets you change parameters later, wire in sliders and do parametric design. The idea is the same as Grasshopper, but there is no Bake: what you see in the 3D viewport is the result.

Opening the panel

• Press Ctrl+` or click "Graph" at the right of the toolbar.
• The panel sits at the bottom, like the terminal in VS Code. Its title bar has "maximize / restore" and "×".
• Drag the divider to resize. Dragging it very low collapses the panel. Double-click the divider to restore the default height.
• When closed, a handle stays at the bottom. Drag it up or double-click it to reopen.
• The open state and height are remembered on this computer.

Basics

• Add a node: type a name in the search box at the top of the panel, or click a category (Input, Curve, Solid, Transform, Analysis, Guide, Output) and pick from the list. You can also double-click an empty spot on the canvas to open the search box.
• Quick slider: type a number in the search box to create a slider. 0<5<10 sets min, value and max at once.
• Wire: drag from an output port to an input port. Ports are colored by type; mismatched types cannot connect.
• Edit parameters: directly on the node card, or in the properties panel in the left column.
• Inspect data: wire a "Panel" node to any output to see its value. "Statistics" shows counts and sums.
• Select and delete: click a node, drag a box, press Delete. Selection in the 3D viewport and in the graph is the same selection.
• Errors: nodes with errors are marked and the toolbar shows the error count. A node missing an input needs that port wired.

Groups, scribbles, notes and disable

• The graph toolbar has Group, Ungroup, Disable, Scribble and Note buttons. Shortcuts work only while the graph has focus (use ⌘ on Mac).
• Group Ctrl+G: wraps the selected nodes in a colored frame. Dragging the frame moves the nodes inside; double-click the title to rename; 6 colors. Ctrl+Shift+G ungroups and keeps the nodes.
• Scribble adds a line of large text on the canvas; Note adds a sticky note. Double-click to edit the text.
• Disable Ctrl+E: a disabled node does not compute and has no output; its card is marked "Disabled". Press again to enable. Pinned elements and the site and code nodes cannot be disabled.
• Groups, scribbles, notes and disabled state are stored in the design, so they are saved, synced to the cloud and shared. Each action is one undo step.

Input, text and geometry components

• Value List: one "name = value" per line; click "Edit items" on the card to change them. Modes are dropdown and checklist. Outputs the value, the name and the index.
• Button: outputs True while held and False when released. Creates no undo step.
• Graph Mapper: remaps numbers through a curve. Curves are Linear, Bezier (default), Sine and Parabola; drag the control points on the card.
• Text: Concatenate, Text Split, Text Join, Format (supports {0}, {1:F2}, {0:N0}, {0:P1} and similar).
• Domains: Construct Domain, Deconstruct Domain, Remap Numbers (with no source domain wired, it uses the minimum and maximum of the numbers; can clamp).
• Ellipse: center, X and Y radius, rotation; outputs the curve, the two foci and the area.
• Voronoi, Delaunay, Convex Hull: take points. Voronoi can take a boundary curve; without one it uses an expanded bounding box. There is a point limit.
• Color by Value: colors solids by a number. Gradients are Spectrum (blue to red), Heat, Green to Red, Grayscale or a custom stop list; the range is automatic by default, or set your own limits.

Planes, facade subdivision, morphs and clusters

• Planes: Construct Plane, Deconstruct Plane, Plane Normal, Plane 3Pt, Plane Origin, Rotate Plane, Plane Offset and the Plane parameter; planes convert to and from the modeler's construction planes. Orient and the morphs transform points, curves, solids and surfaces alike.
• Facade subdivision: build a surface with Ruled Surface, Loft Surface, Extrude Surface or Plane Surface, split it with Divide Domain², then use Isotrim, Surface Frames, Quad or Triangle Panels, or Surface Box.
• Morphs: Twist, Bend, Taper, and Box Morph (fits an object into every twisted box). The axis is a line segment; meshes and curves are subdivided first so large faces can bend.
• Clusters: select nodes and press Cluster on the graph toolbar; outside links become the cluster's inputs and outputs (up to 8 each, defined by Cluster Input and Cluster Output nodes). Double-click a cluster to edit inside; the level bar at the top goes back to the main graph (Esc), and every change inside is undoable. Explode puts the inner nodes back on the graph.
• Cluster ⇄ component: select a cluster and press "Cluster⇄Comp" on the graph toolbar: the cluster's geometry outputs are connected to a component definition and one instance is placed (you can add more instances later, or convert to a family in the modeler); selecting a component node instead packs its definition chain into a cluster (sliders and values become cluster inputs, and all instances still share it). A cluster's "Instance parameters" are its inputs: duplicate the cluster and connect different inputs, and you have different instances of the same definition. See Modeler: Component ⇄ cluster.
• Path Mapper: remaps data tree paths with text masks, for example {A;B} → {B;A}; (i) binds the item index; targets can use integer expressions such as {A;(i/3)}. One mapping per line; branches that match no mask are dropped with a warning.
• Data Dam: holds back upstream changes until you press Release ▶. Modes: Manual release, Auto (release after a delay), Open (pass through). Releasing leaves no undo step.
• MD Slider: drag a point in a square to set two numbers at once; outputs the point and X and Y. Arrow keys nudge, Shift for larger steps.
• Value List modes: Dropdown, Check list, Value Cycle (click for the next item, wrapping around) and Value Sequence (step with the arrows, stopping at the ends).

Cost, quantities and type info (analysis nodes)

• Cost summary: uses the Estimate quantity rules and the unit-price table you import (CSV) to give the total, cost by category (walls, floors, roofs, ceilings, rooms, doors, windows, stairs, structural columns, beams, structural slabs) and cost per element. The “Unit price multiplier” and “Opening deduction threshold” inputs can take sliders. Only your imported prices are used; no official or built-in prices are included. Without a price table the total is 0 and a hint is shown.
• Material takeoff: material areas and volumes from the layer build-ups (same as the Material Takeoff schedule). Filter by category and material keywords (separated by |); outputs totals, per-material subtotals and per-element quantities.
• Type info: connect Types to the BIM types node and give a type id; outputs total and core thickness, layer count, materials and thicknesses per layer, U-value and mass per area (simplified; when a material has no λ or density the output is empty and a warning is shown).
• Analysis nodes run after walls, doors, windows, floors and other elements are computed. While you drag they keep the last result and recompute on release. Their outputs are numbers you can use as optimizer goals or constraints, or wire “Elements” and the per-element values into Color by Value.
• Analysis results cannot feed walls, floors or other elements, levels, types or global parameters (that would be a feedback loop); such links are refused. To change sizes based on a result, read it and adjust by hand.
• EEWH massing estimate: before exterior walls exist, builds an assumed envelope from the automatic mass floors and runs the same EEWH energy estimate as the green building panel. The “Window-to-wall ratio” and “Overhang depth” inputs can take sliders or global parameters (ratio -1 = use the green building panel). Outputs the main indicator (Req for residential, ENVLOAD for air-conditioned, AWSG for schools and large spaces), its limit, indicator / limit (below 1 passes), Uar, Uaw, SF, pass/fail, and the floor, envelope, wall, roof and window areas and shape factor.
• EEWH envelope estimate: once walls, doors and windows are drawn, uses the model envelope and outputs the same indicators plus the actual window-to-wall ratio. With an overhang depth ≥ 0 every wall window uses one uniform overhang; -1 keeps the per-window settings of the model and panel. Both EEWH nodes take building category, county, climate zone and altitude from the green building panel and Project information (the category can also be set on the node); a hint appears when the design has no saved north or county.
• Daylight factor (preliminary): daylight factor on the work plane of every room under the CIE overcast sky. The “Glazing transmittance” and “Grid spacing” inputs can take sliders or global parameters; other settings come from the environment panel. Outputs the area-weighted mean DF, minimum DF, uniformity, % of area meeting the target, number of rooms, rooms meeting the target (count and share), worst room mean and whether all rooms meet the target. The node computes on the main thread with at most 5,000 points and 4 million samples (stricter than the panel).
• Natural ventilation (preliminary): effective ventilation opening ratio of every room (against the panel threshold preset, or a threshold set on the node), cross-ventilation path, wind- and buoyancy-driven airflow. The “Wind speed” and “Wind direction (from)” inputs can take sliders. Outputs rooms, rooms passing (count and share), minimum opening ratio, cross-ventilated rooms, minimum and mean air changes, total airflow, mean indoor air speed and whether all rooms pass.
• Thermal comfort (preliminary): PMV/PPD and adaptive comfort. Inputs: air temperature, mean radiant temperature, relative humidity, air speed, clothing, metabolic rate, outdoor monthly mean; unconnected inputs use the panel (the outdoor temperature uses the hottest filled month). Outputs PMV, PPD, operative temperature, |PMV| ≤ 0.5, comfort temperature, whether the wide/strict acceptability range is met and whether the outdoor temperature is within the model range.
• Structure summary: the Structure panel preliminary sizing and quantities — concrete m³, rebar kg, formwork m², slab area, longest beam span, longest slab short span, column/beam/slab counts, warnings, elements smaller than preliminary and “No warnings”. Recommended sizes are written back only with “Apply all preliminary sizes” in the Structure panel (one undo step); the node never resizes anything by itself.
• Frame analysis (preliminary): a linear-elastic frame analysis summary of the structural columns and beams — maximum storey drift ratio, whether it is within 0.005, maximum demand/capacity ratio, number of elements with D/C above 1, base shear, period, seismic weight, maximum displacement, maximum moment and axial force, and pass/fail. Loads and seismic coefficients come from the structure settings (see Frame analysis). This is a preliminary reference only — not a certified structural design; enlarged sizes are applied only with the button in the Structure panel, and the node never changes elements. With more than 1500 columns plus beams the node does not compute; use the Structure panel instead.
• Parking count check and Accessibility check: the same as the Parking and Accessibility tabs of Egress / Accessibility / Parking Checks; output required and provided numbers, failures per group and pass/fail, usable as optimizer constraints. Thresholds and rules are editable defaults, not a code ruling. Egress (travel distance, two directions) is a background calculation, not a node, and cannot be an optimizer goal.
• All costs and quantities are preliminary estimates, not a formal budget or quantity survey.

Code checks, area and regulation summaries (analysis nodes)

• Site, Rules, Envelope and Code checks are back in the node list under “Analysis” (the Site page in the left column stays off). These nodes in older designs show and compute as before.
• Linked 2D drawing: reads the curves, texts and block insertion points of the chosen layers of a 2D design in the same project (or a 2D design shared with you), the equivalent of Revit's linked CAD and AutoCAD's external references. In the properties panel pick the design and tick layers; the outputs are curves, closed curves, each curve's layer, texts and text positions, and insertion points and block names. Units are converted to metres; “Origin to drawing” uses the snapshot's lower-left corner as the origin (for large coordinates such as TWD97); “Elevation from text” gives curves and insertion points the elevation of the nearest numeric text (for example a contour label 12.5), and “Text search distance” limits the distance. Curves can feed the Property Line node's “Boundary curve” input (the first closed curve becomes the boundary vertices), the toposurface's “Contours”, “Survey points” and “Boundary” inputs, and any node that takes curves, such as walls and grids. The content is stored as a snapshot (inside the design); when the 2D drawing changes the properties panel says so, and only “Update” re-reads it (the cloud has its own button). Objects generated from 3D (imported views, fields) are not read, so 3D reading 2D while 2D imports views from 3D never loops. Limits: 5,000 curves, 50,000 vertices in total, 5,000 texts, 20,000 insertion points, 64 layers; during live collaboration a snapshot over 56 KB cannot be sent, so select fewer layers. AI cannot create this node.
• Property Line → Rules: the Property Line node has a new “Site” output (the property polygon plus the setback you enter for each edge). Connect it to the Site port of Rules to get the BCR area, FAR floor-area cap, effective height and buildable area. Every property-line edge is treated as a neighbour boundary; for road edges and Art. 164 planes use the Site node of older designs.
• Code checks (preliminary): connect an Envelope (with Art. 164 planes and setbacks) or just the Derived output of Rules (BCR, FAR and height only). Outputs: BCR use, FAR use, whether BCR / FAR / height pass, Legal (preliminary), mass floor area and top height — usable as optimizer goals or constraints, or with Color by Value. Several code checks in one evaluation share a single union of the masses; while you drag they keep the last result and recompute on release.
• Area scheme summary (preliminary): reads one area-plan scheme (by name, or the first scheme of the kind) and outputs the counted total, total of all types, building coverage (union of the level projections), ping and area per level. FAR and coverage schemes are compared with a limit (limit input → Rules node → user input of the regulation review report), giving the usage ratio and “Within limit”.
• Sales area summary (preliminary estimate): same as the Sales Area table — main, annex, canopy, common, parking, total sales area, common-area ratio and ping, plus the sales area and ratio per unit.
• Regulation review summary (preliminary): the main figures of the Regulation Review Report plus whether BCR / FAR / height pass, the overall result and the numbers of failing and undetermined items. Items that cannot be determined (missing limit or design value) do not count as passing.
• All of these are preliminary results from the conditions and area schemes you enter, not a review by the authority; their results cannot feed walls, floors or other elements.

Query, phase filter and schedule summary (analysis nodes)

• Query elements (preliminary) (Query): filters model elements by category (walls, doors, windows, floors, rooms, columns, beams… or all), level key, tag / layer and a project parameter (name = value; blank value = the parameter is filled in, case-insensitive). Outputs ids, names, types, categories, levels, parameter values, areas (walls = net area, the same quantity rules as schedules), volumes, plan outlines, BIM specs, and the element count with area and volume totals. Category, level and parameter value can also take inputs (for example the id from a level lookup). By default only the main model and primary design options are read; you can switch to all design options. The lists hold at most 5000 items (the element count is the full total). The Query node outputs no solids, so nothing is duplicated in 3D or in exports. It can also filter by usage, material, phase created and design option (each with its own input port): usage = a built-in name or code (inherited from the upstream geometry source), material = a material of the wall/floor type layers or the object material (case-insensitive, the bim: prefix is optional), phase created = phase id or name, design option = option id or name (main = main model; when set, the main-model-only restriction no longer applies); changing an element’s usage, material or phase recomputes only the query node.
• Phase filter (preliminary) (Phase filter): applies Revit’s phase + phase filter to elements: pick a phase (id or name; blank = the last one) and a filter (Show All, Show New, Show Previous + Demo…), and it outputs the visible element ids with their status (new, existing, demolished), the count per status, total area and volume, and the new volume. Connect the ids of a Query node to look at only those elements.
• Schedule summary (preliminary) (Schedule summary): the start, finish, total duration (calendar days), task count, critical task count, total cost (5D estimate) of the 4D schedule, and the cumulative cost and completion at the “status date”. The duration factor (for example 1.2) and the start date can take a slider or a global parameter; when not connected the 4D panel settings are used. It suits an optimizer objective or constraint (for example duration ≤ 540 days).
• When element properties, tags, names, design options or phases change, only the Query / Phase filter nodes that read those fields recompute; other nodes stay untouched. These three nodes are also analysis nodes: they keep the last result while dragging and recompute on release; results cannot be connected back to walls, levels, type tables or global parameters.

Sun hours (analysis node)

• Sun hours (preliminary): the same computation as the "Sun hours" tab of the 3D green panel (see Sun hours). Choose the day (winter solstice, March equinox, summer solstice or a custom YYYY-MM-DD), the analysis surface (site ground, neighbouring ground, upward faces, facades), grid spacing, margin / ring width, sampling interval and threshold. Outputs the minimum, mean and maximum sun hours, the share of cells below the threshold, the mean sunshine ratio, the shortest longest-continuous sunshine, the number of valid cells and the hours per cell.
• Inputs: latitude and longitude (empty = Project information; without either, the Taipei reference point with a warning), grid spacing and threshold (all can take sliders or global parameters), and a boundary (closed curve; without it the design's property line is used).
• Background computation: first shows "Computing sun hours…"; after a change it shows the previous result until the new one is ready. Each set of settings is computed once and shared with the 3D tab. Keeps the last result while dragging.
• Optimizer: the result is not immediate, so the optimizer blocks it like a script node (it cannot start when the fitness or a constraint depends on a Sun hours node).
• Physical computation only: the threshold is your own input and the node makes no sunlight-regulation determination. Results cannot feed walls, floors or other elements; the AI does not create this node.

EEWH score (analysis node)

• EEWH score (three indicators, preliminary): the same computation as the "EEWH score" tab of the 3D green building panel (see EEWH score). Outputs greenery RS2, site water retention RS3, daily energy RS4 (assessed items), the subtotal of the three and its maximum, TCO2 / TCO2c, λ / λc, EEV, whether the daily energy threshold passes, whether all three could be computed and the number of missing data items.
• Inputs: the node has no parameters; it uses the entries saved in the design from the panel, the building category and county of the green building panel, and the site from Project information and the Rules node. Missing data are listed in the node warnings.
• Only the Taiwan rule pack provides it: with other rule packs every output is empty and the node warns "Not provided by this rule pack". Preliminary, not an EEWH label review, no label grade. It is an analysis node: it keeps the last result while dragging, and results cannot feed walls, levels, type tables or global parameters.

Levels node: unequal storey heights and level lookup

• The Levels node has new inputs “Elevations” (zs) and “Level names”: connect a Series or list to build levels with unequal storey heights.
• Level k keeps the id of the existing k-th level, so walls and rooms do not lose their level when the count changes.
• Connecting a raw number list to a wall’s level input still works but shows a warning (ids follow list positions); connect the Levels node’s level table instead.
• New node “Level at” (LevelAt): elevation or name in; level id, name, elevation, level above and storey height out.
• New node “Level info” (LevelInfo): level count, ids, names, elevations, storey heights, typical storey height and level-table warnings.

Structural grid node: list inputs, intersection planes and grid coordinates

• The Structural grid node has new optional inputs: X/Y spacing lists (xs, ys; a connected Series or slider list replaces the parameter text, at most 60 spans each way, items of 5 cm or less are skipped with a warning), rotation (angle) and origin. Old files without these links are unchanged.
• New outputs: X/Y spacings, X labels (1, 2…), Y labels (A, B…) and intersection planes (one plane per intersection, X axis along the grid, ready for the target plane of Orient).
• Global parameters can drive each span (“X span 2→3” and so on) and the grid rotation: assign them in the structure settings section of the Global Parameters dialog. The saved spacing text stays the same; only that span changes when the model is computed.
• The legacy Grids (Grid) node is marked “Legacy” in the list. Select it and type upgradegrid (upgradegrid all = every one) to upgrade it to a structural grid: equal spacing becomes “n*d”, origin / rotation / overhang carry over, and it stays the same node, so links from its lines and intersections and the intersection order are kept. A grid whose spacing is driven upstream, or with a single line in one direction, cannot be upgraded (the reason is shown).
• The command line and point input accept grid coordinates: B-2, A3, 3-A; A3+1.2,0.5 = offset from the intersection along the grid directions. Examples: line B-2 C-4, cplane origin A1. If the model has no such intersection the text is read as an ordinary coordinate.
• cplane grid A3: puts the construction plane origin at grid intersection A3 with its X axis along that grid system (rotated grids included).
• “Convert to floors and auto-follow” in the Mass Floors section of the outliner turns mass floors without a floor element into real floors (like FF) that follow the mass: resize the mass and the floor boundaries change in the same edit (one Ctrl+Z); they do not follow while dragging, only when you release. If the face is lost the previous shape is kept with a warning. Off by default; “Stop all auto-follow” turns it off. The Floors node, the area schedule and auto mass floors share one set of sections, so their numbers agree.

Plane follow, reference planes, thickened surfaces and room curves

• A plane that follows a CPlane or a level: on the card of a Plane node, Follow can pick a named CPlane (saved with cplane save) or a level. When that CPlane is overwritten or the level elevation changes, the plane and everything downstream follow live. Picking a level connects the level table of that building’s Levels node to the L input automatically. Offset along normal moves the followed plane further. If the followed CPlane is deleted, the node keeps its stored plane and shows a warning.
• Publish as CPlane: click Publish as CPlane on the card and the toolbar CPlane menu gets a Graph-driven (read-only) group; picking it moves the viewport CPlane to this plane. The entry reads the computed result and is never written to the design’s named CPlane list, so it cannot trigger a recompute in return. Reference Plane nodes have the same option.
• Reference Plane: a project-level reference plane, a dashed line in plan and a vertical plane in 3D. Create one with refplane x1,y1 x2,y2 name or from the node list. When a plane is connected to input P (for example a Plane node that follows a CPlane), the graph places it. For dimensions and global parameters see Modeler › Global parameters.
• Surface to Solid: thickens a Grasshopper surface along its normal into a solid that supports booleans, volumes and IFC or 3DM export. Ruled Surface, Loft Surface and Extrude Surface also gain a Thickness parameter and a Thickened (B) output; thickness 0 = surface only, so older files are unchanged. Thickened shells are components and are not counted as mass.
• Solid Face: picks faces of a mass: vertical faces, upward faces (slopes included) or all. Outputs the boundary curve (its winding is the outward normal), the normal, a plane on the face, a surface for four-sided faces and the area. Index −1 = all faces. The mass stays visible; it is not consumed.
• The Face input of curtain systems and roofs by face: connect the boundary from Solid Face (or any closed curve or Grasshopper surface) to Face and the face is driven live by the graph: when the mass changes, the curtain system or roof follows at once, without Update to Face. One face per curve; several faces of a curtain system form one layout, so schedules and IFC still read it. Elements created by face do not follow automatically by default; connect this input for the ones that should. Older elements without the link keep using Update to Face.
• Structural and architectural hosts driven by the graph: structural columns have a new Points input and beams a new Curves input: connect the intersections or lines of a structural grid (or any point or curve list) for one column per point and one beam per segment (polylines split at their vertices); changing the grid spacing moves them. When one node makes several, element ids are “node#number” and marks get -1, -2…; size and level stay on the node, and Apply recommended sizes in the Structure panel uses the largest recommendation among them. Re-running the automatic layout does not delete these graph-driven columns and beams. Connecting the Boundary curve of a ceiling (a list of closed curves = outer loop + holes), the Profile curve of a shaft (the first closed curve), the Path curve of a stair (two points = straight, more = multi-run along the path, up to 64 points) or the Profile curve of an extruded roof (a polyline drawn in a vertical plane: first to last point = reference line, point heights minus the reference level = profile) replaces the sketched boundary, path or profile. Older elements without a link are unchanged. Up to 500 columns or beams per node.
• Live walls by face (first step): Solid Face has new Base edge (E) and Face height (H) outputs: the bottom edge of each vertical face, directed so that the wall exterior faces out of the mass. Connect Base edge to a wall’s Location curve and set the location line to Finish Face: Exterior; the wall sits on the mass face with its body inside the mass and follows the mass at once. Wall height still comes from the base and top constraints. Limitation: when the number of faces changes (for example a new corner), walls are renumbered and doors or windows may move to another wall; to keep openings in place, use Wall by Face with Update to Face.
• The Curve input of room separation lines: connect Grasshopper or Rhino curves; polylines are split at corners and each segment becomes a separation line. When the curves change, room areas update live.
• Rooms of project: lists every room with its outline, name, number, area, level key and node id; it can list one level only, or enclosed rooms only. It is an analysis node, so its results cannot feed separation lines, walls or other elements.
• Results of analysis nodes cannot be connected to a reference plane, the Face input of a curtain system or a room separation line; such links are refused (they would create a feedback loop).

Merged geometry nodes and dimension references

Loft, sweep, revolve, ruled surface and extrude used to exist separately in SketchUp, Rhino and Grasshopper. The node list now shows only one of each. Old nodes (from old files, or created by the AI under old names) are calculated as before: type and parameters are unchanged, they are just not in the list. Type an old name in the search box (for example LoftSurface, Sweep1Srf) and you find the merged node.

• The "Output" of a newly created Loft node defaults to Solid. Loft Surface nodes in old files have no such parameter and remain Rhino meshes, so their values do not change. The AI node catalog lists only the merged nodes; old names sent by the AI are still accepted.
• 3D dimension references: the "Dimension (3D)" node has two new inputs, "Start reference" and "End reference". When you measure on a vertex of a mass or curve with the SketchUp dimension tool, Rhino dim or the tape measure with D, it connects to that node automatically. The reference records which vertex of that output it is (for example the top of point 2 of the outer outline of prism 0). When upstream changes, the dimension follows; if the vertex disappears, it keeps its last position and a warning appears on the node. Dimensions without a reference use their stored coordinates as before.
• Global parameters created by the tape measure with G go into the "Measure" group. When you measure a vertical edge of an extrusion, a "Read Global Parameter" node is created and connected to the extrusion height. See Modeler.

Code nodes and AI skills

• Collapse to code: box-select a group of nodes and press "To code" on the graph toolbar. The group becomes one code node holding a readable program in a TypeScript subset (one nodes.Rectangle(…)-style call per node). Wires coming in become its inputs; wires going out and geometry without downstream nodes become its outputs (up to 8 each). Results are identical to the original group, and one undo brings the nodes back. Supported for now: Number, Slider, Point, Vector, Series, Range, Remap, Expression, list nodes, Statistics, Line, Polyline, Rectangle, Polygon, Circle, Move, Rotate, Scale and Extrude; other nodes (BIM elements, booleans, meshes…) are listed as "cannot be converted to code yet" and stay as they are.
• Edit the code: select a code node and edit its code and input/output names in the properties panel; it applies on blur or with Ctrl+Enter. Type annotations, type, interface and as are supported; import, class and enum are not. Helpers such as pt(), circle(), extrude() and print() are available.
• Sandbox: the code runs in an isolated Web Worker with no network, no DOM and no access to your account or anything outside its own inputs. Each run is limited to 2 seconds and 8 MB of output; beyond that it stops and says why on the node. The card shows how long the node ran in the sandbox.
• Trust: when you open a design someone shared or that you imported, code nodes do not run by default, and the graph shows "This design contains code nodes. Click to trust and run them". Read the code in the properties panel first. Your choice is remembered on this computer per design and code; if the code changes, you are asked again.
• AI skills: press "AI skill" and describe a rule in one sentence, for example "from the 11th floor up, widen the corridor and mark fire compartments". The AI returns one code node (inputs, outputs and code); read it and press "Add and run". From then on it runs deterministically in your browser without calling the AI again. Only your description is sent to the AI, never your design; check any regulatory values against your local code.
• Skill library: save useful skills with "Save to library" (kept on this computer) and place them in other designs with one click; export and import JSON to share them. Imported skills also need "trust and run" after placing.
• Data trees, meshes and surfaces: group inputs can be data trees (computed per branch, paths kept). Moving, rotating or scaling mesh solids or surfaces, slanted extrusions, and the Extrude surface output and thickened shell are computed by the main program with the original nodes, so results match the group. Inside the code these results can only go on to Move, Rotate, Scale or list nodes, or out of the group. When Extrude produces an empty solid, the code node outputs an empty solid too, like the node.
• Expand back and revoke trust: a code node collapsed from a group whose code has not been edited by hand can be put back into its original nodes with "Expand to nodes" in the properties panel (one undo collapses it again); edited ones explain why they cannot. "Revoke trust" on the card stops that code from running in this design; you can also revoke any trusted code one by one under the home account menu → Settings → Trusted code nodes.

Grasshopper map

This map is maintained by hand. The node list below is generated from the source code.

Node list

Nodes from the command line

The 3D page's command line creates nodes directly. A new node connects to the current selection, for example select a curve and type extrude 12. Type help for every command. The full table is in 3D modeling › Command line.

• rect 0,0 10,8: a 10 × 8 rectangle at the origin.
• extrude 12: extrude the selected curve by 12.
• slider 50 0 100: a slider with value 50, range 0 to 100.
• example flared: load an example (a slider-driven radial flared column).