The library

Primitives

Thirty-two starting shapes, split into fourteen bases to cover and eighteen components to cover them with. Each one is built from parameters rather than loaded from a file, so every one comes with its own controls.

Why a library at all

Blender's own primitives are fine, and mostly wrong for this. A UV sphere crowds every pattern into a knot at its poles. A default cube has six faces. A Suzanne has an ngon count that makes half the tools refuse.

These thirty-two exist because computational design needs surfaces with particular properties: even quads, no poles, known curvature, clean boundaries. They are the shapes that make the rest of Rhea behave.

Every primitive has controls

After adding one, its parameters appear in the Library section — teeth on a gear, turns on a coil, curvature on a saddle. Adjust them and the primitive is rebuilt.

These controls rebuild, they do not deform

Changing a primitive's parameters regenerates the mesh from scratch. If you have already edited or sculpted it, those edits are gone. Set the parameters you want first, then start working on it.

Bases — surfaces to cover

Added at the 3D cursor, or at the origin if you would rather. Every one is quad-dominant and evenly distributed, and every one was chosen because it exposes a different problem: boundaries, poles, double curvature, changing face size.

Controls

Gridcolumns, rows, size: Flat quad grid, the simplest base there is. Use it when you want to see what an effect actually does without curvature confusing the picture.
Blobresolution, radius: Lumpy quad sphere. The default choice for organic effects, because the irregularity gives curvature-driven fields something to respond to.
Torusaround, through, radius, thickness: Closed quad torus with no poles. Every face is a well-shaped quad, which makes it the most reliable test surface in the library.
Tubearound, along, radius, height: Open cylinder for columns and vases. Has boundaries at both ends, so it is a good place to check how an effect handles open edges.
Saddlecolumns, rows, size, curvature: Hyperbolic paraboloid — the gridshell roof form. Curves in two directions at once, which is what makes it the standard test for panelisation.
Vaultarc, length, radius, depth: Barrel vault for arcades and hangar roofs. Single curvature, so every panel on it can be flat, unlike the saddle.
Domeresolution, radius: Even-quad dome with no pole to bunch the pattern. A UV sphere would crowd every effect into a knot at the top; this one does not.
Mobiusaround, across, radius, width: One-sided band with a single edge. Worth a look for what it does to any effect that assumes an inside and an outside.
Helicoidalong, across, turns, height: Spiral ramp — the minimal surface of a helical stair. Twists continuously, so tessellated components rotate as they climb it.
Catenoidaround, along, waist, height: The minimal surface a soap film pulls between two rings. Good starting point for form finding, since it is already close to a solved shape.
Supershaperesolution, radius: Parametric organic solid from the superformula. Produces forms that look designed rather than generated, with sharp lobes and creases.
Torus knotwindings, loops, resolution, thickness: Trefoil swept into a tube — a form that ties itself. Long and continuous, so growth and geodesic fields run interestingly along it.
Seashellresolution, sides, turns, growth: Logarithmic spiral that grows rather than repeats. Face sizes change continuously along it, which makes it the best test of whether an effect scales properly.
Corrugatecolumns, rows, waves, depth: Folded sheet panel, stiffened by its own profile. Flat overall but locally curved, which is a different problem for an effect than either extreme.

Components — shapes to cover with

Components are built to sit inside a unit cell: normalised to a one-by-one footprint in X and Y, with Z running from zero upward from the surface. That is the convention Tessellate expects, and it is why these drop straight in without scaling.

Controls

Gearteeth, depth: Toothed disc, the classic tessellation test piece. Teeth from three to thirty-two.
Hex tiledepth: Hexagonal plate for honeycomb panels and faceted cladding.
Ringsides, hole, depth: Annulus for chainmail and perforated screens. The hole control goes to nearly half the radius, which is thin enough to read as wire.
Scaledepth: Overlapping shingle for scales, roof tiles and armour. Set tessellate Scale slightly above 1 so neighbours overlap.
Spikeheight: Four-sided spine for spike fields. Combine with Taper for something sharper than the primitive alone.
Weavedepth: Arcing strand for woven and knitted patterns. Needs a clean quad grid underneath or the over-under reading breaks down.
Brickdepth: Chamfered block for walls and greeble. The chamfer is what stops it reading as a plain cube.
Findepth: Upright blade for louvres, gills and fins. The component that most rewards the Tilt control.
Starpoints, depth: Star plate for decorative screens. Three to sixteen points.
Studsides, radius, height: Round boss for rivets, pins and panel fixings.
Frustumbase, top, height: Flat-topped pyramid — blunt where the spike is sharp. Set the top to zero and it becomes a spike anyway.
Trussreach, width, depth: Diagonal cross brace for diagrids and space frames. The structural component of the set.
Petalsmoothness, depth: Pointed leaf blade for petals, feathers and scales with a soft edge.
Gemfacets, table, girdle: Faceted crystal for jewelled and cut-stone panels. Table at zero gives a point rather than a flat top.
Archsmoothness, depth: Archway member for arcades and vaulted screens.
Podsides, rows, height: Swelling bulb for seeds, buds and growths. The most organic component, and the one that responds best to Randomness.
Coilsides, turns, wire, height: Helical spring for threads and wound patterns. Half a turn to twelve.
Twiststeps, twist, width, height: Square post twisting as it rises. Has its own twist independent of the tessellate Twist control, so the two compound.

Using your own component

Any mesh works. Rhea normalises whatever you give it into the unit cell, so a component that is a hundred units tall and offset from the origin still lands correctly. For predictable results:

Model it sitting on the XY plane, growing upward in +Z, rather than centred on the origin.
Apply its scale and rotation. Non-uniform object scale is the single most common cause of a tessellation that looks subtly wrong.
Keep the vertex count honest. Every vertex is multiplied by the face count, and a component you thought was cheap at four hundred vertices is four million over ten thousand faces.
If you want welded seams, make sure the component's boundary vertices sit exactly on the cell edges. They cannot weld to their neighbours if they do not meet them.

Pairings worth trying

Saddle + Truss. A real gridshell. Take it through form finding and then fabrication for the full architectural pipeline.
Seashell + Scale. Because face sizes change along the spiral, the scales grade in size on their own without any field at all.
Torus knot + Coil. A wound thread following a form that already ties itself. Absurd and worth seeing.
Corrugate + Fin. Louvres on a folded panel, the two rhythms interfering with each other.
Catenoid + Ring. A perforated minimal surface, which is a surprisingly good stress test of boundary handling.