An agent device includes a microphone which collects audio in a vehicle cabin, a speaker which outputs audio to the vehicle cabin, an interpreter which interprets the meaning of the audio collected by the microphone, a display provided in the vehicle cabin, and an agent controller which displays an agent image in a form of speaking to an occupant in a region of the display and causes the speaker to output audio by which the agent image speaks to at least one occupant, and the agent controller changes the face direction of the agent image to an direction different from an direction of the occupant who is a conversation target in a case that an utterance with respect to the face direction is interpreted by the interpreter after the agent image is displayed on the display.