How likely are third parties to intervene in repeated conflicts between adversaries? Can third party intervention redirect competitive interactions toward collectively beneficial cooperation? Does mutual cooperation persist when the third party can no longer intervene in the conflict? To address these theoretically and practically important questions, we introduce the Repeated Peacemaker Game. Adversaries and third parties interacted for 60 game rounds in our incentivized experiment, producing a rich dataset of 7200 decisions. Participants made decisions under one of two conditions: In the Early Intervention condition, third party intervention was initially possible and then became impossible. In the Late Intervention condition, third party intervention was initially impossible and then became possible. Third parties in the Early Intervention condition intervened at high rates and established cooperative norms that outlasted the intervention period, resulting in sustainable cooperation even after intervention was no longer possible. By comparison, third parties in the Late Intervention condition intervened at relatively lower rates, but still effectively promoted cooperation, despite the initial history of competition between adversaries. We find strong evidence for ‘conditional cooperation’, whereby adversaries and third parties reciprocate each other’s choices in preceding rounds: Adversaries were more likely to cooperate when third parties intervened in the preceding round; third parties, in turn, were more likely to intervene when one or both adversaries cooperated in the preceding round than if both adversaries competed in the preceding round. These findings help explain when, why, and how third parties promote cooperation in groups.