How Qatalyst works.
An agentic pipeline that handles the full optimisation workflow, from natural-language problem statement to solver output with explanation.
Intake and formulation
Describe your problem. Qatalyst classifies it (VRP, TSP, knapsack, scheduling, general binary), extracts decision variables, constraints, and objective, and builds a formal optimisation model.
Solver chooser
A hybrid rule-engine plus LLM selects the best solver for the problem size, structure, and time budget. Auto-tunes simulated annealing parameters.
Classical and quantum
Every problem runs on a classical solver by default. When it fits quantum hardware, Qatalyst reformulates to QUBO and routes to an annealing backend.
Validation
Solutions get checked against the original constraints, scored for quality, and either pass, flagged for review, or sent back for retry.
Solver backends
| Backend | Type | Status |
|---|---|---|
qatalyst_sa | Classical simulated annealing (own IP) | Live |
mock_quantum | QUBO mock (dev and test) | Live |
| D-Wave Leap | Quantum annealing | Planned |
| Gurobi | Exact MILP | Planned |
Want to try it on your data?
Join the waitlist and tell us about your problem.