// the find
reinterpretcat/vrp
A Vehicle Routing Problem solver
A Rust implementation of the Vehicle Routing Problem with support for rich VRP variants — time windows, multi-day routes, reloads, driver breaks, skills, job priorities, and more. It uses a custom metaheuristic called Rosomaxa (a GSOM-based population strategy) rather than off-the-shelf algorithms. Aimed at developers who need a serious VRP solver they can embed or extend, not a black-box SaaS.
The constraint/objective extension model is genuinely well-designed — you can compose custom VRP variants from building blocks rather than forking the solver. Python bindings via PyO3 mean you can benchmark it against OR-Tools or Google's CP-SAT without writing Rust. The Rosomaxa algorithm is documented with internals exposed (the experiments directory even has visualization tooling), which is rare for a metaheuristic implementation. Scientific benchmark formats (Solomon, TSPLIB, LILIM) are supported out of the box, so validating solution quality against published benchmarks isn't a manual exercise.
The author is explicit that this is a pet project with 'permanently experimental' status and 'very limited support' — adopting this in a production logistics system means you own it. There's no built-in routing engine; you supply a pre-computed distance/time matrix, which means you need to wire in OSRM or Valhalla yourself before the solver is useful on real maps. The pragmatic JSON format is custom and underdocumented in places — the error index in the docs exists but error messages in practice often just give you a code. No HTTP API or streaming solution updates, so long-running solves block synchronously.