ecosystem-primer

Evaluate requirements in order: use Deep Agents if you need planning, file management, subagents, or persistent memory; LangGraph for custom control flow like deterministic loops and branching; LangChain create_agent for single-purpose agents with a fixed tool set.

LangChain is the framework layer providing model abstractions, tools, and the create_agent loop. LangGraph is the runtime layer underneath it, offering durable execution and explicit StateGraph control flow with nodes, edges, and conditional edges.

Set LANGSMITH_API_KEY, LANGSMITH_TRACING=true, and LANGSMITH_PROJECT to enable tracing. These are the current variable names; older names no longer work. Model provider keys like ANTHROPIC_API_KEY or OPENAI_API_KEY are set separately as needed.

Yes, the layers compose. A compiled LangGraph graph can be registered as a named subagent inside Deep Agents, and the orchestrator delegates to it via the task tool. LangChain tools and retrievers work inside both LangGraph nodes and Deep Agents tools.

If the LangChain Docs MCP server is connected, query it with tree, cat, and rg commands against docs.langchain.com paths. Otherwise fetch https://docs.langchain.com/llms.txt to find relevant pages, then fetch those pages directly.

Avoid Deep Agents when the task is simple enough for a single-purpose agent with fixed tools, or when you need precise hand-crafted control over every graph edge. In those cases use LangChain create_agent or LangGraph directly.