Agent Execution And Reasoning Loop

via “agent loop orchestration with middleware and tool execution”

Framework for building LLM applications with chains, agents, retrieval, and tool use.

Unique: Integrates with LangGraph to implement ReAct agent loops with middleware support for step interception and custom execution logic. Agents are composable Runnable objects that support both sync and async execution with built-in state management and tool call marshaling.

vs others: More flexible than rigid agent frameworks and more observable than black-box agent implementations because middleware enables custom step handling and state inspection at each reasoning/acting cycle.