Cacharpo is an autonomous agent that serves as an assistant within live coding performance. Cacharpo uses techniques of Machine Listening and Music Information Retrieval to “hear” musical parameters made by a human live coder and respond to them by typing code and making music. The development of Cacharpo addresses the challenge that typing as a physical act can take excessive time undermining the theatricality of the performance. Cacharpo’s main focus is music influenced by cumbia sonidera, a genre developed in Mexico and derived from Colombian cumbia. Cacharpo consists of four subsystems: a low-level feature extractor, an Artificial Neural Network to identify cumbia roles and other musical parameters, an algorithm to generate SuperCollider code, and SuperCollider classes providing notations for economical live coding of cumbia sonidera elements.