Where to Run the LLM Agent Harness: Sandbox vs. Local Architecture Tradeoffs

Two architectures for running agent harnesses, the tradeoffs between them, and how we make skills and memories work when the harness isn't local.