LangGraph

What it does

• Low-level orchestration for stateful, multi-step agents — graphs of nodes with explicit state channels
• First-class checkpointing: persist state, resume, rewind, and time-travel through agent runs
• Human-in-the-loop interrupts, streaming, and durable execution built into the runtime
• LangGraph Platform for managed deployment; LangSmith for tracing
• Python and JavaScript parity (pip install langgraph / npm install @langchain/langgraph)

When to use it

When you outgrow prompt-and-loop agents: workflows that need resumable long runs, auditable state transitions, or human approval gates. Deliberately more verbose than CrewAI — you trade boilerplate for control.

Getting started

# Python
pip install -U langgraph

# JavaScript / TypeScript
npm install @langchain/langgraph

Things to know

'Draw your own graph' means a real design phase — budget for it. Claude runs through the langchain-anthropic package (init_chat_model), which you install and configure separately. The LangSmith/Platform pull can be more ecosystem than a solo project needs.