The Sm–Nd isotope systematics and geochemistry of eclogitic, websteritic and peridotitic garnet and clinopyroxene inclusions together with characteristics of their corresponding diamond hosts are presented for the Letlhakane mine, Botswana. These data are supplemented with new inclusion data from the nearby (20–30 km) Orapa and Damtshaa mines to evaluate the nature and scale of diamond-forming processes beneath the NW part of the Kalahari Craton and to provide insight into the evolution of the deep carbon cycle. The Sm–Nd isotope compositions of the diamond inclusions indicate five well-defined, discrete eclogitic and websteritic diamond-forming events in the Orapa kimberlite cluster at 220 ± 80 Ma, 746 ± 100 Ma, 1110 ± 64 Ma, 1698 ± 280 Ma and 2341 ± 21 Ma. In addition, two poorly constrained events suggest ancient eclogitic (> 2700 Ma) and recent eclogitic and websteritic diamond formation (< 140 Ma). Together with sub-calcic garnets from two harzburgitic diamonds that have Archaean Nd mantle model ages (TCHUR) between 2.86 and 3.38 Ga, the diamonds studied here span almost the entire temporal evolution of the SCLM of the Kalahari Craton. The new data demonstrate, for the first time, that diamond formation occurs simultaneously and episodically in different parageneses, reflecting metasomatism of the compositionally heterogeneous SCLM beneath the area (~ 200 km2). Diamond formation can be directly related to major tectono-magmatic events that impacted the Kalahari Craton such as crustal accretion, continental breakup and large igneous provinces. Compositions of dated inclusions, in combination with marked variations in the carbon and nitrogen isotope compositions of the host diamonds, record mixing arrays between a minimum of three components (A: peridotitic mantle; B: eclogites dominated by mafic material; C: eclogites that include recycled sedimentary material). Diamond formation appears dominated by local fluid–rock interactions involving different protoliths in the SCLM. Redistribution of carbon during fluid–rock interactions generally masks any potential temporal changes of the deep carbon cycle.