Platform

One live model, re-solved whenever it changes

Routite is six layers that turn planning from a nightly batch into something that runs all day. Each layer can be adopted on its own, but they are built to run as one loop.

The route through the platform

  1. IngestOrders, GPS, traffic, constraints
  2. TwinLive network state
  3. OptimizeGPU re-solve
  4. ForecastETA and disruption
  5. ConsoleCopilot and approvals
  6. ExecuteDriver apps, TMS, WMS

Stop 01 · Architecture

Six layers, from feed to driver

Status as of November 2026. Nothing here is in production use yet.
LayerWhat it doesBuilt onStatus
1 · IngestPulls orders, telematics, traffic, depot and warehouse state, constraints and SLAs into a feature store.RAPIDS cuDFIn build
2 · Live twinKeeps a continuously updated model of the network, fleet and orders.Network graphIn build
3 · OptimizationWarm-starts cuOpt from the current plan and re-solves on change or on cadence.NVIDIA cuOptIn build
4 · Forecast and disruptionPredicts travel, service time and demand; flags deviations worth a re-solve.Triton, TensorRTIn build
5 · ConsoleCopilot questions, what-ifs, approvals and guarded execution.NeMo, GuardrailsIn build
6 · Execution and feedbackPushes plans to driver apps, TMS and WMS; real outcomes retrain models.ConnectorsPlanned

Stop 02 · Flywheel

Every day of driving improves the next plan

Plans are only as good as their travel and service-time estimates. Routite records what actually happened at every stop and feeds it back.

More days of history give better ETA and demand models. Better models give plans that need fewer mid-day changes. Fewer changes mean dispatchers trust the plan, so more of the network runs through it.

1

Drive

Edge logs arrival, service and departure at every stop.

2

Learn

ETA and service-time models retrain on your own outcomes.

3

Plan

The next solve starts from better estimates.

4

Trust

Fewer surprises, so more routes run through Routite.

Stop 03 · Deployment

Run it where your data has to live

Routite cloud

Managed

We run the GPUs and the stack. Fastest way to start a pilot.

Your VPC

Your cloud account

The same stack deployed into your cloud account, on GPUs you control. Planned.

On-premises

NIM and NVIDIA AI Enterprise

For operators who cannot send network data out. Planned, sized per customer.

Stop 04 · Control plane

What your operations team configures

ConstraintsTime windows, capacities, skills, driver hours, vehicle types, zones.
ObjectivesDistance, lateness, overtime, emissions, or a weighted mix.
Re-solve rulesCadence, and which events trigger an immediate re-solve.
Approval rulesWhich changes need a dispatcher and which can go through.
RolesViewers, dispatchers, planners and administrators.
ConnectorsWhere orders come from and where routes are written back.

Stop 05 · Security

Guarded by design

Executed changes are logged with who approved them, what moved and why. Copilot cannot act outside the permissions of the person using it.

SOC 2 and ISO 27001 are on the roadmap and are planned, not held. Details on the trust and security page.

  • Role-based access per depot and region
  • Audit log of every executed change
  • Data residency by region (planned)
  • Single sign-on with SAML or OIDC (planned)

Stop 07 · Destination

Next step

See the platform run on your own data

A replay uses a past day of your orders and vehicles. No live connection, no change to how you dispatch today.