A key challenge in integrated natural resource management (INRM) consists in reconciling ecological and socio-economic dynamics for a collective, sustainable, and equitable management of the land. Diverse stakeholders have different legitimate points of view and representations of the system they exploit and its management is better seen as a collective learning process. We have tested a "companion modeling" (COMMOD) approach combining multi-agent systems (MAS) models and role-playing games (RPG) to examine concrete INRM problems with stakeholders. This participatory process consists in repetitive back and forth steps between the model and the field to facilitate dialogue, shared learning, and collective decision-making through interdisciplinary and action-oriented research. The methodological features of this approach are de-scribed and two applications in rural Thailand are presented. Among the lessons learnt, we argue that COMMOD users need to distinguish between two specific contexts: (i) The production of new knowledge on a given complex system and (ii) Supporting evolving, iterative, and continuous collective decision-making processes. Such MAS-based and bottom-up COMMOD approach offers a great potential where land management policy frameworks encourage the decentralized management of natural re-sources.