Choose a deployment model
Choose direct hosted workspace use, a hosted API, a private service or local Python execution.
Work directly in a hosted workspace
A hosted application gives planners a browser interface for preparing inputs, configuring supported manifest settings, running a solve and reviewing the result. You do not need to build a custom UI to begin a provisioned evaluation.
Routing workspaces can bring jobs, fleet resources and route inspection together. Rostering workspaces can show a weekly plan, coverage, availability and assignment diagnostics. The browser uses its application backend; solver credentials do not belong in page JavaScript.
Request a walkthrough to confirm access, enabled problem families, data import, result export and operational review for your workspace. This documentation site is not a solver session or self-service signup.
A provisioned service
Your backend submits structured requests to a separate solver endpoint. This separates application releases from solver operation and lets the deployment own worker placement.
Agree supported problem types, capacity, region, data handling and service terms during provisioning. This site does not promise a public self-service endpoint or a universal SLA.
A private service
A service inside your controlled infrastructure can expose the same integration boundary while letting your team operate the runtime and networking.
Size the worker for the selected engine. In the server/worker topology, the worker owns accelerator execution. Health of the API process alone is not proof that an accelerator is ready.
An embedded process
A supported local Python entry point runs in the process you manage. You own environment preparation, dependency pinning and execution limits.
Local execution does not automatically establish an air-gap guarantee. Validate packaging, license requirements, telemetry settings and every configured dependency against your network policy.
Evaluate with your workload
Compare feasibility coverage, objective quality, wall-clock time, memory and operating cost on the same dataset. Include preprocessing, data transfer and startup in the measurement that matters to your application.
CPU and GPU trade-offs depend on the model and instance size. Hardware alone does not establish a performance guarantee.