Engineering Dynamics Company

Point Cloud Toolkit™

Point Cloud Processing & Visualization for Blender

Point Cloud Toolkit — User Guide

Importing, filtering, and surfacing laser-scan point clouds in Blender.

This guide walks through installing the add-on, finding its panels, and completing the common tasks, with a reference for every control. For a terse feature list, see README.md.

Contents

1. Overview
2. Installation
3. Choosing which tools appear
4. Quick start
5. Import Point Cloud
6. Filter Point Cloud
7. Create Ground Surface
8. Create 3D Surface
9. Documentation & provenance report
10. Optional dependencies
11. Tips & troubleshooting
12. License & support

A note on values: defaults are shown in parentheses. Distance controls follow the scene's unit system in Blender; their defaults are given here in metres, Blender's base unit (1 unit = 1 m).

Overview

Point Cloud Toolkit adds an EDC Toolkits tab to the 3D View sidebar (press the N key) with four tool sections that share one point-cloud selection, plus a display section for the viewport:

The add-on targets Blender 4.x and is numpy-vectorized throughout, so it stays fast on million-point clouds.

Installation

1. In Blender, open Edit → Preferences → Add-ons → Install… and select the PointCloudToolkit-<version>.zip file.
2. Enable Point Cloud Toolkit in the add-ons list.
3. Open the 3D View sidebar with the N key and select the EDC Toolkits tab.

The add-on's preferences let you rename the sidebar tab — set it to an existing tab's name to share a tab with other add-ons.

On Blender 4.2+ the same zip also installs as an extension (Preferences → Get Extensions → Install from Disk), and updates automatically when added from the EDC Software repository.

Choosing which tools appear

If you only use part of the toolkit, you can hide the rest. Open Edit → Preferences → Add-ons, expand Point Cloud Toolkit, and use the Features checkboxes:

All four are on by default. Unchecking one removes its panel from the sidebar immediately — no restart needed. Documentation always stays visible, so the user guide and provenance report remain reachable.

Nothing is uninstalled: the tools are only hidden, and re-checking a box brings the panel back with its settings intact.

Note. The Filter settings still apply when you build a surface, even with the Filter Point Cloud panel hidden — subsampling and outlier removal run as part of Create Ground Surface and Create 3D Surface. Hide the panel only if you're happy with the current filter settings.

Click Save Preferences (or enable Auto-Save Preferences) so your choices persist the next time you open Blender.

Quick start

1. In Import Point Cloud, click Import Point Cloud and load a scan. The imported cloud is selected automatically as the source for the other sections.
2. *(Optional)* Set Clip To Object to restrict surfacing to a boundary mesh.
3. *(Optional)* In Filter Point Cloud, enable Subsample (Voxel) and/or Remove Outliers (SOR), then click Filter → Create New Point Cloud.
4. For drivable ground, use Create Ground Surface; save the .blend first so the texture JPG can be written next to it.
5. For vertical or overhanging geometry, use Create 3D Surface (Poisson is the all-rounder). Afterwards, Trim Bulges (To Cloud) removes invented geometry and Rebake Texture re-fits the texture.

Import Point Cloud

Imports PLY (ASCII or binary), PTX, PTS, E57, LAS/LAZ, and PCD files as a coloured mesh point cloud displayed through a Geometry Nodes modifier (also available on File → Import). The imported cloud is automatically selected as the Point Cloud source used by the other sections.

Every imported cloud gets its own Geometry Nodes display tree; the controls for it are in Point Cloud Display, immediately below.

• PTS is plain ASCII and declares nothing about its own columns, so the layout is read from how many there are: x y z, x y z intensity, x y z r g b, or x y z intensity r g b. RGB is taken as 0–255 unless the values are already within 0–1. A scan with intensity but no colour is shown as greyscale, stretched over that file's own intensity range — PTS does not say whether an intensity is -2048…2047, 0–255 or 0–1, and stretching is the only reading that works without being told. It affects display only. The leading point-count line is used where present and not required, since exporters differ; several scans concatenated into one file are read as one cloud.
• Ill-formed files. A file that cannot be parsed is reported as an error naming the file, and the import is cancelled; the full detail goes to the System Console. Nothing is half-imported.

Points whose coordinates are not finite (NaN or Inf — truncated writes, a scanner error row) are dropped, and the count is logged. This matters more than it sounds: the check for "is this cloud far from the origin?" is a maximum over the coordinates, so a single NaN made it false and a georeferenced scan imported unrecentered at 500,000 m — losing exactly the precision recentering exists to protect, silently. A single Inf was worse: it made the centroid infinite, so every point in the file came out unusable.

• Source Units / Scale — pick the units the file is stored in (Meters, Feet, US Survey Feet, Inches, cm, mm) and the importer sets Scale to convert the cloud into metres (Blender's 1 unit = 1 m). Choose Custom to type any multiplier yourself. Scale is applied to the coordinates on import.
• Global Shift (Auto) — georeferenced scans (UTM, state plane) store coordinates in the millions, where Blender's float32 mesh storage loses precision and the cloud looks jittery. Shifting the cloud back near the world origin fixes that; the removed offset is recorded on the object (custom properties pct_global_shift / pct_import_scale) so the original coordinates can be recovered — the same idea as CloudCompare's Global Shift. Three modes:
  • Auto (default) — recenter only when the coordinates are very large (subtracts the rounded centroid); ordinary local clouds are left in place.
  • None — do not shift at all; import at the file's true coordinates.
  • Manual — subtract a Manual Shift X/Y/Z you enter (e.g. your own false origin) instead of the automatic value. The value is in scene metres (after Source Units/Scale), so for a cloud already in metres it is just the source offset. It is stored on the object like the auto shift, so coordinates still round-trip.

Knowing which frame a cloud landed in. Two scans of one site overlap only if they share a shift, so the import says which happened rather than leaving it to the console:

• The Import Point Cloud panel, and the file browser's options under Auto, show the actual offset that will be used and which cloud it came from — before you import, while it can still change your mind. Three states read differently: nothing imported yet; an import that needed no shift (named, so you can tell it from the first case); and an offset in play, with its numbers. Forget Remembered Shift clears it so the next cloud sets its own.

A cloud already near the origin — anything local, or a georeferenced one scaled down from centimetres — is moved by nothing, so it leaves no offset to inherit. That is not the same as not having imported, and it is not the same as the offset being gone: if an earlier cloud did set one, it stays in play and the panel says so rather than naming the wrong cloud.

• The status line after an import reads either *"Aligned with the earlier import"* or *"New global shift …, remembered for the next import"*.
• The flag is stored on the object (pct_shift_reused) and appears in the Provenance Report as *Aligned with earlier import*.

Manual Shift X/Y/Z is only shown in Manual mode. It is only read there, and showing it under Auto made it look like a value that would be used.

Importing several clouds that share a georeference: the shift applied to the first cloud is remembered on the scene (pct_session_shift), and Auto reuses it for later imports as long as that offset keeps the new cloud near the origin — so multiple scans of the same site import aligned to each other (like CloudCompare reusing one global shift for a session), instead of each getting its own centroid. A cloud from a different, far-away site still gets its own shift. Each cloud also records its own offset in the custom properties pct_global_shift / pct_import_scale, so file = (local + global_shift) / scale recovers the originals.

• Import progress (file read, point count, mesh build) prints to the system console — opened automatically on Windows, or the launching terminal elsewhere.
• PLY, PTX, PTS, uncompressed LAS, and ASCII/binary PCD are read natively. E57 needs the pye57 package and LAZ needs laspy[lazrs] — both install automatically the first time such a file is opened, or via the Install E57 / LAZ Support Now button. Compressed PCD uses Open3D.

Point Cloud Display

A second view of the selected cloud's Geometry Nodes modifier: it mirrors every input the display tree declares, labelled as the tree labels it, in the tree's own order — the same rows the Modifier Properties panel shows, in the sidebar where you are already working. Nothing here touches the data; filtering, surfacing and baking always use every point, whatever the viewport is showing.

As imported, that is:

• Points Visible % (100) — thins the viewport display to keep it responsive on dense clouds. Lower draws fewer points and navigates more smoothly; 100 shows all.
• Point Radius — the displayed size of each point, in scene units. Raise it to make a sparse cloud read as a solid surface, lower it for crisp detail.

Because the panel mirrors rather than matches by name, editing the display tree is reflected straight away: add an input and it appears here, remove one and its row goes with it. Values are per-cloud — each import has its own display tree, so adjusting one never changes another.

Which cloud these edit. The active object, when it is a cloud — unlike the processing tools below, which follow the pinned Point Cloud. The two rules disagree as soon as a second cloud is imported, because importing pins the new one: selecting the first and moving a slider would have changed the second, which looks exactly like the panel not working. Display is about the thing you are looking at, and the Showing: line names it. The pinned cloud is still the fallback, so the panel is not empty when something else is selected.

This section is not one of the four tools, so it stays available even with every tool hidden in Preferences. A cloud too dense to navigate is a problem you have before you reach any of them.

How this works, and why it needed to. Blender offers no way for a Python panel to draw a geometry nodes modifier input — its own Modifier Properties panel is built in C, and layout.prop() needs a struct and a property name. Blender 5.x keeps each socket's value inside an ID property group, which prop() refuses as its data argument outright.

So each row here is one of the add-on's own properties, wired to the selected cloud: reading it reads the modifier, and setting it writes the modifier. Nothing is stored on the side, so the panel and the Modifier Properties panel cannot drift apart — there is one copy of the number and it lives on the modifier. Up to eight inputs are mirrored, which is well past the two the display tree declares; an input the modifier holds no number for (a Geometry input, or one removed from the tree) gets no row.

Where the value physically lives has moved between releases — 3.x and 4.x use an ID property on the modifier, 5.x an entry under modifier.properties.inputs — so the add-on finds it rather than assuming, and Diagnose Display Storage prints the whole shape to the System Console if it ever cannot.

If the rows are missing. A cloud imported by an older version, or one whose modifier never had a value written for an input, gives the panel nothing to mirror and the rows do not appear. Restore Display Controls writes the values back onto the selected cloud's modifier — taking each input's own default from the node group, or the import defaults (Points Visible % 100, Point Radius 0.01) when the group has none — and the rows come back. It only fills inputs the modifier is actually missing, so a value you have already set is never overwritten, and inputs that hold no number (a Geometry input) are left alone. If an input still cannot be placed, the System Console names it; Diagnose Display Storage is the next step.

Create Ground Surface

Drapes a regular grid over the cloud to build a drivable ground surface. Distance controls are given in metres (1 unit = 1 m).

• Resolution (Cell Size) (0.3 m ≈ 1 ft) — spacing of the surface grid. Smaller means finer detail, more polygons, slower; larger is coarser and faster. Set it to the smallest real feature you need to keep. A single surface is capped at 8 million grid vertices; if your cell size and scan extent would exceed that, a dialog appears pre-filled with the smallest cell size that fits. You can accept it, type your own cell size (the dialog shows the resulting grid live as you change it), or Cancel and clip to a smaller area to keep the finer resolution. If a value you enter is still too dense it is raised automatically to fit.
• Grid Angle (Z) (0°) — rotate the surface grid about the vertical axis so its faces run along this direction instead of world X/Y. Line it up with a roadway that isn't axis-aligned to get clean, rectangular faces along the road (and fewer of them for the same detail). The surface is created rotated by this angle about the cloud's centre; Rebake Texture, tiling, and the aerial tools all follow the rotation automatically. 0° keeps the world-aligned grid.
• Ground Percentile (10) — the height percentile taken as "ground" within each cell. Lower samples from below, hugging the lowest returns and rejecting overhead noise (trees, wires, vehicles); raise it only if the true ground is sparse and stray low outliers are dragging cells down.
• Below-Grade Reject (0.5 m) — blanks any cell sitting more than this far below its neighbours (a stray below-ground return) and refills it from the surrounding ground. Lower rejects more aggressively; 0 disables.
• Spike Reject (Above) (1 m) — removes upward spikes where trees, poles, or walls leave a cell with no ground returns (so even the low percentile lands high). Any cell rising more than this above the local ground level is refilled from the surrounding ground. Lower removes shorter bumps too (may cut real curbs); 0 disables.
• Spike Width (Max) (6 m) — the widest above-ground object Spike Reject may remove; anything broader is kept as real grade (hills, embankments). Raise to clear wider vegetation strips; lower to protect narrower real rises.

This is the setting that decides whether a wood comes out. Spike Reject compares each cell against a local ground reference built by opening the grid at this width, so an object *wider* than the setting carries the reference up with it and is read as grade — the spikes stay, and no amount of lowering Spike Reject (Above) will touch them. A belt of trees is routinely 20–40 m across, so a scan full of vegetation spikes usually wants Spike Width raised to comfortably exceed the widest continuous stand, not Spike Reject lowered. The cost is that real rises narrower than the setting are flattened too, so it wants to exceed the vegetation and stay under the terrain: a wide belt of trees on genuinely rolling ground is the case where clipping the cloud beats tuning either number.

• Fill Holes (on) — interpolate empty grid cells from their neighbours so sparse spots don't leave gaps.
• Max Fill Distance (2 m) — how far filling reaches across a gap. Larger bridges bigger holes (but can invent surface over true voids); 0 = unlimited within a connected region.
• Color Height Tolerance (0.25 m) — only points within this height of the sampled ground contribute colour to the surface and its texture, so vehicles and foliage above the road can't tint it. Raise to admit more (risks pulling in above-ground colour); 0 uses all points.
• Texture Resolution (4096) — longest side, in pixels, of the baked colour texture, sampled directly from the cloud so it can be sharper than the surface grid. Higher is sharper, larger, slower; 0 matches the grid resolution. The JPG is saved next to the .blend — save the .blend first.
• Texture Color Source (optional) — sample the texture's colour from a different object (e.g. the original full-resolution cloud) instead of the surface's point cloud. Use it when the geometry cloud is a filtered copy but you still want a full-detail texture.
• Texture Tiles (1) — with N > 1, the ground surface is split into up to N×N tiles, each with its own baked texture, for more total resolution than one big image. It is *up to* N×N because tiles that fall on empty area (holes, or corners the scan doesn't reach) are skipped — so a 4×4 split of an irregular scan may yield fewer than 16 tiles.
• Join Tiles Automatically (on) — after tiling, the tiles are merged into a single object, each tile keeping its own texture as a material slot, so you get the tiled resolution with one object to manage. The shared tile borders are welded (coincident vertices merged), so the joined surface is continuous and a Decimate modifier won't tear gaps along the seams. Turn it off to leave the tiles as separate objects. (Blender's Separate → By Material splits a joined surface back into its tiles.)
• Join Tiles Into One Object — a manual button that appears when two or more separate surface tiles are selected; it performs the same merge (including welding the shared borders), for when Join Tiles Automatically was off or you split a surface yourself.
• Create Ground Surface — builds the surface, bakes the texture, and assigns UVs and an image-texture material. Grid cells with no data (holes the fill couldn't reach) leave no stray points: vertices that aren't part of any face are removed automatically, so you get only the surfaced geometry.
• Rebake Texture (Selected Surface) — shown when a created surface is selected; re-bakes at the current Texture Resolution without rebuilding the mesh.
• Decimate Mode (Un-Subdivide) + Add Decimate Modifier — shown when a mesh is selected; adds a live Decimate modifier to every selected surface to cut polygon count. Un-Subdivide (default) reverses subdivision, removing edge loops a set Un-Subdivide Iterations number of times (each pass roughly quarters the face count) — ideal for the regular ground grid. Collapse merges edges until only a fraction of faces remain, set by Decimate Ratio (0.5 = keep half) — better for the irregular 3D reconstruction. Once added, the panel shows the modifier's Resulting faces count; the modifier stays live so you can adjust or remove it, and Blender's modifier Apply bakes it in.

Aerial image (replace the point-cloud colour)

You can drape one or more aligned aerials / orthophotos over a created surface instead of the colour baked from the point cloud — useful when a recent aerial reads better than the scan colours. Both textures are kept, so you can switch back at any time. This works on ground and 3D surfaces.

1. For each aerial, add a reference and align it over the surface in top view: the simplest is a Plane (Add → Mesh → Plane) given the aerial as an image-texture material, scaled and positioned to cover that aerial's area. (An Empty → Image reference also works.)
2. Select every aerial plane, then the surface last so it is active. (For a single aerial you can instead set Aerial Reference to the one plane, and optionally Aerial Image to override the image.)
3. Click Apply Aerial Image. Each surface face is textured by the aerial whose footprint covers it, with top-down planar UVs computed from that plane's world extent, so every orthophoto lands by geographic position. Because these are real textures (their own UV layer and one material per aerial), they export cleanly.
   • Overlaps: where two aerials cover the same spot, the one with the smaller footprint wins (usually the more zoomed-in / higher-detail image).
   • Gaps: faces that no aerial covers keep their point-cloud texture, so partial coverage just works — aerial where you have it, scan colour elsewhere.
4. Show Point-Cloud Texture / Show Aerial Image toggles the whole surface between the two — the original point-cloud texture (and, for a joined surface, each tile's own texture) is preserved.

If the cloud was recentered on import, aligning the aerials visually over the surface already puts them in the right (shifted) space — no coordinate math needed. Tip: if your aerials are just tiles of one orthomosaic, merging them into a single image first (e.g. QGIS / gdal_merge) lets you use one plane instead.

Tuning & troubleshooting (ground surface)

Adjust one setting at a time and re-create the surface to see its effect.

Create 3D Surface

Full 3D reconstruction via Open3D (installed automatically on first run) for vertical and overhanging geometry — walls, embankments, vehicles, structures. For drivable ground, Create Ground Surface remains the better tool. Distance controls are given in metres (1 unit = 1 m).

• Method — Poisson (the all-rounder; smooth and watertight, but invents geometry past the data — pair it with Density Trim / Trim Bulges), Ball Pivoting (interpolating; keeps the original points as vertices, good for evenly dense clouds), or Alpha Shape (Delaunay concave hull; simple and fast). The controls below change with the method.
• Poisson Depth (9) — octree depth. Higher captures finer detail but is slower and can amplify noise; lower is smoother and coarser.
• Density Trim (0.05) — fraction of the lowest-density (invented) Poisson vertices to cut away. Higher trims more of the ballooned surface (but can eat thin real areas); 0 keeps everything.
• Alpha (0 m = auto) — Alpha Shape wrap radius, in metres. Larger is smoother and bridges gaps; smaller wraps tighter with more holes; 0 auto-picks from the average point spacing.
• BPA Radius Multiplier (1.0) — scales the Ball Pivoting ball radii (derived from point spacing). Raise it to span wider gaps and close holes in sparse areas; too large bridges fine detail.
• Use Existing Normals (on) — when the imported cloud already carries per-point normals (many terrestrial-scanner PLY / E57 / PCD exports do), reconstruct with them and skip estimation and orientation — usually the slowest phase on large clouds. The checkbox is greyed out when the selected cloud has no stored normals, or when voxel subsampling is on (averaging invalidates them); in those cases normals are estimated as usual. Turn it off to always re-estimate.
• Normal Neighbors (k) (30) — neighbours used to estimate normals (only when estimating). Higher gives smoother normals; lower is more local and noise-sensitive.
• Normal Orientation — Consistent (accurate but single-threaded, so slow on big clouds) or Up (near-instant, great for terrain but can misorient steep walls/overhangs). Only applies when estimating.
• Normals are cached automatically. When the toolkit estimates and orients normals (the slowest phase, especially with Consistent orientation), it keeps the result. Reconstructing the same cloud again reuses it and skips that phase — so tweaking Method, Poisson Depth, Density Trim, alpha, Texture Tiles, UV, or texture size and rebuilding is fast. The cache is keyed by the exact points fed to estimation plus Normal Neighbors (k) and Normal Orientation, so it is reused as long as none of those change; editing the cloud, changing a filter (subsample / outlier removal) or clip, or changing k or the orientation recomputes it. The reuse works within a Blender session for any cloud (including subsampled/clipped ones); for a whole-cloud run it is also saved in the .blend (a pct_computed_normal attribute on the cloud) so it survives save and reload. The system console prints whether a run reused the cache or recomputed.
• Texture Resolution (4096) — longest side of the baked colour texture, sampled from the cloud. Higher is sharper, larger, slower; 0 matches the grid.
• Texture UV — Smart UV (default; handles arbitrary 3D shapes, each surface getting its own texture area) or Planar (top-down projection; zero packing waste, best for pure terrain seen from above).
• Texture Tiles (1) — with N > 1, split the surface into N×N tiles, each with its own texture, for more total resolution.
• Join Tiles Automatically (on) — after tiling, merge the tiles into a single object, each keeping its texture as a material slot. Turn off to keep separate tile objects.

Post-build tools (shown when a reconstructed surface is selected)

These appear only when a built 3D surface is the active object. After a build the cloud is often what is selected, and the panel says so rather than simply not showing them.

• Trim Bulges (To Cloud) — deletes surface farther than the Trim Distance (0 m = auto, from point spacing) from the source cloud, removing Poisson's invented geometry. Larger distance keeps more (trims less); smaller trims more aggressively.
• Trim Long Triangles + Max Edge Length — deletes faces with an edge longer than the threshold: the webbing Ball Pivoting, Alpha Shape and Poisson stretch across holes, doorways and scan shadows.

This is a different fault from a bulge, and Trim Bulges cannot reach it. A triangle spanning a hole has all three corners on real scanned points, so measuring vertices against the cloud finds nothing wrong with it. Its *edges* are what give it away.

Max Edge Length at 0 measures the surface itself — a low quantile of its edge lengths, times a wide factor — so the same setting works whether the scan is in metres or feet, and however dense it is. A low quantile rather than the median on purpose: once webbing is a large fraction of the surface, the median lands *between* the good and bad triangles and the automatic threshold floats out of reach of both. Type a length when you want to be precise; the operator reports how many faces it removed, and refuses rather than emptying the object if every face exceeds the threshold. Unlike Trim Bulges it needs no Open3D.

• Rebake Texture (Selected 3D Surface) — re-bakes at a new resolution, sampling the cloud directly.
• Split & Bake Tiles — re-splits the selected surface into Texture Tiles.
• Join Tiles Into One Object — with two or more surface tiles selected, merges them into a single object that keeps each tile's texture as its own material slot. Separate → By Material reverses it.
• Decimate Mode + Add Decimate Modifier — adds a live Decimate modifier to the selected surface(s) to reduce polygon count. Choose Un-Subdivide (set the iterations) or Collapse (set the ratio); the panel then shows the resulting face count. Adjust or Apply it in the modifier stack.

Tuning & troubleshooting (3D reconstruction)

Poisson is the usual starting point; change one control at a time and rebuild.

Documentation & provenance report

The Documentation panel has an Open User Guide button (this guide) and a Provenance Report (Active) button. The report writes a self-contained HTML document for the active cloud or surface — the source file and its SHA-256 hash, point/geometry counts, coordinate extents in both Blender and original (georeferenced) space, the import scale and global shift, and the tool/Blender versions — for documentation and chain-of-custody. It's saved next to the .blend (or a temp folder if unsaved) and opened in your browser. The "Prepared by" field is left blank for you to fill in; it is not taken from the OS login.

Optional dependencies

Optional Python packages install themselves into Blender's Python when first needed — no manual pip work is required. An internet connection is needed the first time each feature is used.

Tips & troubleshooting

• Texture stayed inside the .blend? The JPG is written next to the saved .blend. If the file isn't saved yet, the texture is packed into the .blend instead — save the .blend and rebake (or re-create the surface) to write the JPG to disk.
• Huge cloud is slow? Lower Points Visible % for display, and enable Subsample (Voxel) before surfacing. Textures still bake from the full cloud.
• Grid too large to build a ground surface? A single surface is capped at 8 million vertices. Accept the auto-fit dialog to raise the cell size, type your own cell size there, or clip to a smaller area with Clip To Object.
• Surface looks jittery / stepped, or a warning about being far from the origin? A georeferenced cloud imported with Global Shift = None keeps huge coordinates, and Blender's float32 mesh storage then loses ~0.25–0.5 m of precision. Re-import with Global Shift = Auto (or Manual) so the cloud sits near the origin. Surfacing such a cloud shows a warning for this reason.
• Poisson surface balloons past the scan? That's Poisson inventing geometry where there are no points — raise the density trim, or run Trim Bulges (To Cloud) after the build.
• Progress seems stuck? Long operations print progress to the system console (the terminal Blender was launched from, or Window → Toggle System Console on Windows).
• Install of Open3D / pye57 / laspy fails? Check the internet connection and that Blender can write to its own Python; as a fallback, run pip install <package> with Blender's Python and restart Blender.
• E57 import reports a Visual C++ runtime error? The E57 reader's system libraries need a newer Microsoft Visual C++ runtime than some Blender installs ship. Install the latest Microsoft Visual C++ Redistributable (x64), restart Blender, and retry. Converting the file to PLY or LAS also works.

License & support

Point Cloud Toolkit is © Engineering Dynamics Company, written by Anthony Cornetto, and is free software licensed under the GNU General Public License, version 2 or later — see the bundled LICENSE file. Official builds, updates, and support are provided by Engineering Dynamics Company; visit edccorp.com or contact EDC support.