Skip to main content
V3 is in beta. General availability (GA) is planned for 1 December 2026.
An HVAC contractor in Antwerp has an 11-hour work order: an overhaul of an air-handling unit at an office in Leuven. One technician with the hvac skill must keep the order until it is done. It is longer than one shift. V3 serves each job whole, on one shift. To spread an order over several days, send it as equal pieces, one job per piece, and link the pieces with relations. The solver then decides how many pieces each day takes. The response excerpts on this page are real output for the requests shown. Travel figures can change with the road data.

The scenario

All times are local time in December 2026, written with the +01:00 offset. The drive takes about 51 minutes each way. A technician who leaves at 08:00 arrives at 08:51 and must stop by about 16:10 to be back by 17:00. After the lunch hour, that leaves just over six hours on site per day.

Equal pieces, not fixed blocks

A block sized without knowing the drive leaves part of the day unused. The same order sent two ways, with the same shifts, the same relations and break_interruptible on every job: A 4-hour block and the 3-hour block do not fit together in one six-hour day, so each block takes its own day. One-hour pieces save the third day’s round trip. Pieces at one location add no travel between them, and each day pays setup_duration_s once, not once per piece. See Travel and service time.

How each requirement maps to the request

  • Pieces. Eleven jobs, ahu-overhaul-01 to ahu-overhaul-11, at the same location, with service_duration_s: 3600 and required_skills: [{ "name": "hvac" }] each.
  • Lunch. break_interruptible: true on each piece, so a piece that runs into the lunch break pauses and resumes at 13:00. See Breaks during a long job.
  • One technician. A same_vehicle relation over all pieces. See Relations.
  • No other jobs in between. A consecutive relation over all pieces, in order. It applies within one route: each day’s pieces run back to back.
  • In order across days. One time_lag relation per neighbouring pair, from the earlier piece’s departure ("on": "departure") to the next piece’s start. Piece 7 is never served on a day before piece 6.
  • Whole or nothing. An all_or_none relation over all pieces: if the shifts cannot hold the whole order, every piece stays unassigned. Leave it out to allow a partial order.

Solve the order

Read the response

The response has two routes, both for ines. joris is unused because same_vehicle keeps the order with one technician. Piece 4 runs into lunch. It starts at 11:51 and ends at 13:51, and its stop reports the paused hour in break_s. See Breaks in the response.
To show the order as one activity per day, merge each route’s run of pieces, from the first piece’s service_start to the last piece’s departure.

Choosing the piece size

  • Smaller pieces fit each day more exactly. One-hour pieces suit most days. Use 30-minute pieces when days are tight, such as a long drive or other jobs before and after the order, for orders of up to 8 hours.
  • At most 16 pieces in a time_lag chain. The time_lag relations link the pieces into one group, and a linked group holds at most 16 tasks. For a 24-hour order, use pieces of at least 1.5 hours. Without the time_lag relations a chain may be longer, but its pieces are then ordered only within each day. See Limits.
  • Every piece is a task. Splitting multiplies the task count against the 400-task solve limit, so a week of orders for a team often needs an async run.
  • Each extra day costs a round trip. The solver favours fewer, fuller days, but a day may still take a single piece when that is cheapest. To make each extra day cost more, set costs.per_route. See Vehicle count is free until you price it.

Next steps

Field service day

A single day with skills, loads and lunch.

The V3 API model

Every request and response field.