Your harness, your memory

Agent harnesses are the dominant and necessary scaffolding for building modern AI agents by orchestrating interactions between LLMs and external tools. These harnesses are intrinsically linked to agent memory, as they manage context, state, and how an agent learns from user interactions. Using closed, proprietary harnesses, especially those behind APIs from model providers, means relinquishing control over this crucial memory, which creates platform lock-in. This prevents developers from switching models and gives up the proprietary data flywheel that memory creates, ultimately advocating for open-source harnesses to ensure data ownership and model optionality.