Agent composition

What happens if you put DSPy inside the RLM sandbox REPL? So instead of just doing subLLM calls with (prompt, context) params the agent can send ({signature, modules}, context) and compose its sub-agents more predictably

View original