We present a technique that enables a software agent to augment its ontology with domain related concepts by collaborating with other agents. The collaborating agents have their own individual ontologies, they can share concepts and relationships that relate to a requested speci?c concept (which is known as a fragment). Thus, speci?cally, our technique selects the fragments that will be shared. This approach enables agents to answer queries with more range and detail, and it also enables an agent to infer new exploitable knowledge. Without this capability, an agent may be limited by its domain model, and cannot re?ect changes in the environment. Through empirical evaluation, we show that our technique reduces the cost of acquiring concepts that are regularly used (compared with learning nothing) and reduces the complexity of the agent?s ontology by augmenting it with selected concepts and relationships which are related to its domain (compared with learning everything).