Prime Agent: A self-improving RLM agent

Prime Agent is a new open-source self-improving coding agent that uses two core abstractions—Recursive Language Model (RLM) and Continual Harness—to enable dynamic adaptation and learning. Unlike traditional agent harnesses with fixed tools and static configurations, Prime Agent allows models to programmatically manage their own context, sub-agents, prompts, and memory through a persistent IPython kernel, treating these elements as modifiable components rather than fixed design choices. The system is designed to work with frontier AI models and can be used as a general coding assistant, autonomous evaluator, and research collaborator.