Routing for delivery agents and dispatchers

Multi-stop optimisation, OD matrices, and corridor search for the agents and backends that move things.

Logistics is going agentic. Customer service bots quote ETAs. Driver assistants reroute when an address fails. Backend orchestrators feed live GPS into route optimisation between every drop. The shape of the problem hasn't changed. The thing calling the API has.

Footstep is built for that. Routing responses carry explicit units (distance_meters, duration_seconds), turn by turn legs, and a terrain block, so an LLM can answer 'is this a sensible rural delivery route?' without a follow-up call. Geometry is opt-in to keep agent context clean.

Backends get the same data over plain HTTP. The agent surface and the dispatcher view share one source of truth and one set of response shapes.

Patterns

What this looks like in practice

Common shapes mapped to the tools you'd reach for.

Multi-stop optimisation

Pass 8 to 30 stops. Get back the optimised order, the savings versus the input order, and per leg distance and duration. The cost surface respects road network and travel mode, so the optimised tour reflects real driving time, not great-circle distance.

OD matrix for dispatch

Compute an N by M travel-time matrix between sources (warehouses, depots) and targets (customers, drop points). Use it for assignment, ETA quotes, or live re-planning when a driver becomes available.

Corridor search

Find fuel, charging stations, food, or rest stops along an existing route, ranked by how little detour they add. Tunable via max_detour_seconds.

Geocode and route in one call

When the destination arrives as free text ('the next stop is the Asda on Bromley Road'), call find_and_route. Geocoder confidence and up to three alternate matches come back so the agent can offer choices.

Tools you'll reach for