Every source against every target in one request: distance and duration for each pair, by car, bike or on foot. The question behind dispatch, store finders and delivery slots, answered as one table instead of a hundred routes.
10,000 units free every month. No credit card, no sales call.
| min | Jordaan | De Pijp | Oost | Wester. | Zuidas | IJburg |
|---|---|---|---|---|---|---|
| ACentraal | 7 | 14 | 13 | 8 | 24 | 33 |
| BMuseumplein | 10 | 5 | 14 | 14 | 12 | 37 |
| CAmstelstation | 22 | 11 | 11 | 25 | 18 | 28 |
Highlighted: the nearest courier for each order.
POST /directions-matrix/v1 · 3 sources · 6 targets18 pairs · one request · 18 units
Nobody ships an endpoint. They ship a dispatch board that knows who is closest, a store finder that sorts by minutes rather than kilometres, a checkout that promises a slot it can keep. This is what a matrix looks like from the outside.
Three couriers, six orders, one request. Each order goes to the nearest courier by bike.
One request returns the time from every courier to every order. Sort each column and the assignment falls out. No routing loop, no waiting on twenty-four separate answers.
The same road graph, routedThis is the same 3 × 6 table, read the way a dispatcher reads it: an order, the nearest courier, the minutes. Every number here is from the response, not typed in.
Then route the winnerA dispatcher’s morning, one request each time
1 couriers × 2 orders
2 pairs, 1 unit each
A matrix costs one unit per source-target pair, the same as one route per pair. The difference is that it comes back as one table in one round trip, up to 2,500 pairs at a time.
See what a month costsconst { sourcesToTargets } = await rijwind.matrix({ sources: couriers, targets: orders, costing: 'bicycle', }); // nearest courier per order const pick = orders.map((_, j) => argmin(sourcesToTargets.map((row) => row[j].time)));
Sources, targets and a profile go in; a row per source with a cell per target comes out, each cell carrying seconds, distance and the indexes of the pair. Unreachable pairs are null, not an error.
Read the quickstartThe response is the table you asked for: a row per source, a cell per target, each cell with seconds, distance and the indexes of the pair. Your inputs come back with it, so a row never gets matched to the wrong courier.
Everything a developer scans a product page for, in one place, without the adjectives.
Numbers as of September 2026. The pricing page has the full ladder.
A table of travel times and distances from a set of sources to a set of targets, computed in one request. Three couriers and six orders give a 3 × 6 table; each cell says how long that courier would take to reach that order. It is the input to dispatch, store finders, delivery-slot checks and anything else that starts with 'which of these is closest'.
The numbers are the same as you would get from one route per pair. The difference is one round trip and one table instead of dozens of calls, and no route geometry you would only throw away. When you then need the actual turn-by-turn route for the winning pair, ask the routing endpoint for that one.
One unit per source-target pair, from the same monthly pool as maps, geocoding and routing. A 3 × 6 table is 18 units; a 10 × 10 table is 100.
Up to 2,500 pairs per request, so 50 × 50, or 1 × 2,500, or anything in between. Larger jobs are split into several requests.
Car, bicycle and pedestrian, each with its own rules. The matrix on this page is by bike; the same points by car had one target no car can reach, which comes back as null rather than as an error.
No. Durations come from the road graph (speed limits, road class, surface and turn costs), not from live traffic. Rijwind Traffic is a separate live layer for your map and does not feed into these numbers today.
A row per source with a cell per target. Each cell carries the time in seconds, the distance in kilometres or miles, and the source and target indexes so you never have to trust the order. Unreachable pairs have null time and distance.
Yes. The free plan includes 10,000 units a month with no card and no sales call, which is a 10 × 10 matrix a hundred times over. It is a hard cap, so a runaway loop cannot produce a bill.
In the European Union, end to end. No part of the request path touches a US-controlled provider, which is what makes the GDPR and data-residency conversation short. We publish a DPA and a full sub-processor list.
Turn-by-turn for the pair the matrix picked.
The best order for your stops, solved for you.
Everything reachable within N minutes, as a polygon.
Turn the addresses in your table into coordinates.
One unit pool, one table, a cap you set.
What actually differs, feature by feature.
Sessions, units and where the data lives.
Ten thousand units a month, free. The same key covers maps, search and routing when you get to them.