The paper proposes a gossip algorithm-based architecture for managing power flows on radial distribution grids applying decentralized management of demand. Demand flexibility is associated with a cost-like utility function expressing the inconvenience caused by curtailment. Every node of the distribution grid participates in power flow management by sending locally measured and calculated values to its neighboring nodes following a peer-to-peer architecture. The proposed algorithm is based on the application of gossip algorithms to estimate locally critical physical quantities by reaching a global consensus. More specifically, the solution implements gossip algorithms to achieve consensus in aggregated demand and in minimum cost for curtailment of flexible loads/increase of controllable distributed generators.