prime-agent

Why we included this project

Prime Agent is built for work that keeps going after you close the terminal. Its main tool is a persistent IPython session, so file edits, shell commands, and context management all happen through code rather than one-off chat turns. The recursive language model abstraction treats subagents as function calls: you spawn real child agents with rlm(...) and collect their results programmatically, which makes parallel or background subtasks feel like ordinary programming. A continual harness stores memories, skill descriptions, and subagent specs as durable state the agent can refine with evidence-backed updates, and daemon-backed sessions survive disconnects so you can reattach later. For long-running coding and research work, that is the difference between an agent that forgets everything overnight and one that picks up where it left off.