Africa holds immense potential for solar energy, thanks to its high year-round solar irradiation. Advances in photovoltaic (PV) technology and declining costs have made solar energy viable across the continent. However, fluctuating solar irradiance (SI), caused by factors like humidity, temperature and cloud cover, poses challenges for PV systems, causing power quality issues. This could be mitigated by utilising accurate SI forecasting, for optimal integration and operation of PV systems. Very short-term SI forecasts can play a crucial role in minimising energy storage requirements, enhancing power scheduling, and stabilising energy supply in real-time. In Africa, where weak grids usually coincide with abundant solar resources, such forecasts are especially valuable for improving load matching and grid reliability. Despite this, SI forecasting remains limited due to scarce historical SI data and high equipment prices. To address these challenges, the Karlsruhe low-cost sky imager (KALiSI) has been developed for approximately €500, for very short-term SI forecasting. Five KALiSI systems have been deployed in Africa, with data from one system installed in German being used in the present work to train a deep learning model, based on convolutional neural network – long short-term memory (CNN-LSTM) to predict SI. The model consistently delivered lower error rates across different forecast horizons compared to persistence, achieving average normalised root mean square error of 35% at 30 min horizon compared to persistence (50%). Future efforts will adapt this model to the African sites, utilising localised data to refine SI and subsequently PV power predictions, enhancing robustness and accuracy under diverse climatic conditions.