Exactly. One bead at a time - we thought for a while about making it possible to send batches, but would require deep changes to the infrastructure. Also,
in general when doing ab initio it is better to use a separate ab initio code for each bead, as it extrapolates WFC much better in time than across the ring polymer.
Finally, please use the -dev version of i-PI, we have diverged so much from v1.0 that we'll not be able to help you much if you need something. A v 2.0 is
in the making.