// the find
VROOM-Project/vroom
Vehicle Routing Open-source Optimization Machine
Vroom is a C++20 VRP solver that handles TSP, CVRP, VRPTW, PDPTW, and multi-depot variants in a single engine, returning solutions in milliseconds via heuristics and local search. It sits between the routing engine (OSRM, Valhalla, ORS) and your application, exposing either a CLI, an HTTP wrapper via vroom-express, or a C++ library. Aimed at logistics companies and developers who need real-time re-optimization without paying for a cloud routing API.
The algorithm depth is serious — it implements SWAP*, Or-opt, cross-exchange, route-split, and a Christofides-based TSP heuristic, all benchmarked against academic instances with published results. The cost matrix abstraction means you can swap routing backends or supply your own matrix, which matters when OSRM topology doesn't match your business constraints. Active maintenance by Verso with CI across multiple compilers and regression checks against benchmarks on every release is rare for OR tooling. The skills and time-window modeling (per-vehicle, per-job, driver breaks, multi-metric capacity) covers the real-world complexity that toy solvers ignore.
It's a solver, not a service — you still have to stand up OSRM or Valhalla and wire it together, which is a non-trivial operational burden before you get a single route. The HTTP API is a separate Node.js repo (vroom-express), so the C++ core has no native HTTP server; if you want embedded use in a non-C++ stack you're calling a subprocess or going through the HTTP layer with the overhead that implies. Solution quality is heuristic — there's no optimality guarantee, and on pathological inputs the gap to optimal can be material; the README doesn't discuss this honestly. Documentation lives in a wiki that's partly outdated relative to the JSON API spec in docs/API.md, which creates confusion about which source is authoritative.