Open access (OA) is a contested term with a complicated history and a variety of understandings. This rich history is routinely ignored by institutional, funder and governmental policies that instead enclose the concept and promote narrow approaches to OA. This article presents a genealogy of the term open access, focusing on the separate histories that emphasise openness and reusability on the one hand, as borrowed from the open-source software and free culture movements, and accessibility on the other hand, as represented by proponents of institutional and subject repositories. This genealogy is further complicated by the publishing cultures that have evolved within individual communities of practice: publishing means different things to different communities and individual approaches to OA are representative of this fact. From analysing its historical underpinnings and subsequent development, I argue that OA is best conceived as a boundary object, a term coined by Star and Griesemer (1989) to describe concepts with a shared, flexible definition between communities of practice but a more community-specific definition within them. Boundary objects permit working relationships between communities while allowing local use and development of the concept. This means that OA is less suitable as a policy object, because boundary objects lose their use-value when ‘enclosed’ at a general level, but should instead be treated as a community-led, grassroots endeavour.