Dispatch the best driver. Keep the delivery promise.

From the order to the doorstep, one engine.

Each job below maps to one shipped tool: the same routing engine prices the dispatch decision, the round and the detour.

Dispatch

Pick the right driver with a matrix

One many-to-many matrix call answers which driver reaches the pickup soonest: the workhorse behind every dispatch decision. The dispatch card here is a mock-up; the three ETAs in it are real engine output for those three courier positions, and self-hosted there is no per-element metering on recomputing the board.

Rounds

Plan rounds that respect reality

Multi-vehicle optimisation plans stops against time windows, capacities and vehicle profiles in one call. If the van is a 3.5-tonne with a tail lift, the round is planned for that van, not an idealised car.

Optimisation docs →

Coverage

See your delivery zone honestly

Travel-time zones show what a promise actually covers, computed on the real road network, not a circle drawn on a postcode. This is a real eight-minute scooter zone from a Soho kitchen: the river, one-ways and closed streets all shape it.

Along the route

Every detour has a price

Search along a route returns stops with their real engine-computed detour cost, so adding a pickup to a live round is a number, not a guess.

the whole stack, inside your own network

$ docker compose up -d --build

→ tiles · geocoder · router · guidance · territories

→ 0 external map dependencies

Self-hosted

Your order flow never leaves your network

Delivery telemetry is commercially sensitive: where your couriers are is your order book. Self-hosted, none of it transits a third party, and the live-feed architecture proven on thousands of London buses carries courier positions the same way.

Agent-first

Dispatch tools your agents can call

Matrix, optimisation, routing and along-route search are all callable from AI agents through the hosted MCP server, with a card-free API key in one call. The dispatch copilot your team keeps talking about is an afternoon's work, not a platform migration.

Known limit: ETAs come from the routing engine's speed model, not a live-traffic feed, and the optimiser runs as a configurable sidecar. Treat ETAs as what the road network supports, and pair them with your own service-time data.