langchain-oss-primer

Choosing between LangChain and LangGraph depends on your control-flow needs. Use LangChain for standard QA, RAG, and SQL agents, while LangGraph fits stateful deterministic graphs requiring complex orchestration, memory, and persistence management.

You need Deep Agents instead of LangGraph when your stateful orchestration requires managed planning, file handling, and built-in memory. Deep Agents provide higher-level managed middleware compared to LangGraph's deterministic graph control flows.

The right LangChain archetype for a RAG agent with tool integration is the standard LangChain framework layer. It maps directly to QA, RAG, and SQL agent patterns before routing to specific implementation skills for your data tasks.

Setting up dependencies for LangGraph workflows requires confirming your environment expectations first. Match your use case to stateful deterministic graph patterns, then load the specified next skills to begin implementation and subagent delegation.

LangChain can handle subagent delegation and basic memory tasks, but LangGraph is better suited for stateful memory persistence and deterministic orchestration. Select LangGraph when your architecture requires explicit state management and complex control flows.

LangGraph limitations for managed planning include requiring manual graph configuration for file handling and middleware. Deep Agents offer a managed alternative with built-in planning, files, and memory, avoiding the need for custom deterministic graph orchestration.