Cebra is a machine learning tool that creates high-performance latent spaces using non-linear techniques from joint behavioural and neural data. It offers neural latent embeddings for hypothesis testing and accuracy validation, making it suitable for single or multi-session datasets without labels. It also enables the rapid decoding of natural movies from visual cortex and supports the mapping of complex kinematic features. Cebra's code is available on GitHub, making it a valuable tool for neuroscientists seeking to understand adaptive behaviours.