A runtime for a long-lived LLM agent with ambient continuous self-perception, persistent memory, defined authority, domain-specific autonomy, and forensic accou

A runtime for a long-lived LLM agent with ambient continuous self-perception, persistent memory, defined authority, domain-specific autonomy, and forensic accountability, all in a long ongoing relationship with a human. I call this type of system an Artificial Retainer, a non-human cross between a guide dog and someone like your accountant or lawyer. It is not designed to be your friend, but it could be a valuable colleague. Think of this as an attempt to build a trusted stable agent with a stable character that could last decades. https://github.com/seamus-brady/springdrift https://github.com/seamus-brady/springdrift https://arxiv.org/abs/2604.04660 https://arxiv.org/abs/2604.04660