Agroforestry can generate multiple ecosystem services underpinning human wellbeing while maintaining environmental integrity but rather than promoting silver bullet technologies across large areas, menus of options need to be tailored to local context. There are knowledge gaps in scientific understanding about how to enhance tree cover to deliver a range of ecosystem services. We present insights on how to do this from participatory action research across contrasting farming contexts in sub-Saharan Africa. In all cases, farmers had knowledge of a wider range of tree species than was available scientifically or promoted by extension services. This knowledge can be used in tree planting or regeneration initiatives to include a diverse range of tree species, largely unknown to science but important in farmers’ practice. Doing so results in agroforestry options that are more inclusive, with different options suiting different people. Drawing from these experiences we propose a co-learning framework that integrates multiple knowledge systems through the facilitation of stakeholder engagement to identify agroforestry options for different contexts, alongside interventions required in the enabling environment for them to be realised. Co-learning continues beyond the design phase to drive iterative evaluation and refinement of options based on their real world performance, thereby reducing risks and increasing benefits for smallholder farmers adopting innovations.