Local mode is the default mode of execution for graphs. It is also called pseudo-distributed mode, as it emulates distributed mode execution within a single JVM. Being constrained to a single machine, local mode does not scale out, although it can scale up.

Because all partitions in local mode reside within the same JVM, data exchange between partitions can be done entirely in memory. This means that graphs executed in local mode often can be completed in a single phase, repartitioning data in a streaming fashion.