3D bin packing FAQ: short answers, the long one a link away.
Every answer here restates something the site already says elsewhere - nothing new is claimed on this page.
What is 3D cartonization?
Deciding which items fit into which containers, and exactly where, in three dimensions - the problem Packvium solves as a request/response contract rather than a UI.
Request and result contractWhich languages does Packvium support?
Four implementations, written separately: Python and PHP as reference engines, an independent Rust core with a stable C ABI, and a deterministic JavaScript fallback. They reach you as packvium on PyPI and Packagist, packvium on crates.io, @packvium/engine on npm for Node.js, and @packvium/browser - the Rust core compiled to WebAssembly - in the browser. One conformance suite holds all four. Go, Java and .NET are served by ports of the reference engine, on pkg.go.dev, Maven Central and NuGet: they implement the 1.3 contract and are released on their own version line.
PackagesIs Packvium free and open source?
Yes - MIT licensed, across every package in the suite.
ProjectWhat shapes can it pack?
Rigid cuboids, and two shapes a caller opts into per item: a convex hull, decided by exact integer geometry rather than by its bounding box, and a compressible item, whose height falls under load until a declared crush limit. Concave meshes, soft bags and cylinders are still rejected explicitly as items rather than approximated. A container, from 1.6.0, can be a flat mailer that admits a load when its length and width, each plus its thickness, fit.
Item shapesIs the packing guaranteed optimal?
Only where it can prove it. 3D bin packing is NP-hard; Packvium runs a bounded search under an operation-counted budget. Under every one of the seven objectives (the last two, shipping_cost and lowest_landed_cost, from 1.11.0) a result carries the engine's lower bound and its gap to it, and is labelled optimal only when it reaches that bound.
The objective vectorIs Packvium fast?
Measured, and no claim follows from it. Every engine was timed in equivalent containers on one pinned host - cold and warm, with peak memory and every raw artifact retained. The measurement does not support calling Packvium the fastest, so no such claim appears here.
BenchmarksHow do I know the benchmark numbers are trustworthy?
Every number on the benchmarks page is recomputed by an independent validator from the placements each engine actually returned, never taken from an engine's own self-report.
BenchmarksHow do I get started?
Install the package for your runtime and send one request - see Getting started - or try a real request in the Playground first, without installing anything.
First packed containerCan I add custom rules or scoring?
Yes, through five in-process extension points. A rule that must hold for every caller belongs in the request schema instead, so every engine and the validator enforce it the same way - see Extending an engine.
Extending an engineHow do I report a bug, a security issue, or contribute?
A reproducible fixture is the most useful bug report - see Contribute. Report a security vulnerability privately - see Security - never through a public issue.
Contribute