LangGraph

Build stateful agent workflows with LangGraph.

When to Use

Use this skill when the user asks to:

• "setup langgraph"
• "create agent workflow"
• "build graph"
• "state machine"
• "langgraph agent"
• "agent with tools"
• "human-in-the-loop"
• "langchain graph"
• "persist agent state"
• "build multi-step LLM workflow"

What It Covers

1. Install and configure LangGraph
   • pip install langgraph langchain langchain-openai
   • Setup environment variables (OPENAI_API_KEY)
2. Define state schemas
   • TypedDict for structured state
   • MessagesState for conversational agents
   • Custom reducers for state updates
3. Create nodes and edges
   • StateGraph builder
   • Conditional edges (routing)
   • Entry points and END nodes
4. Build agent with tools
   • Tool binding to LLM
   • ToolNode for execution
   • ReAct loop pattern
5. Persist state with checkpoints
   • InMemorySaver for dev/testing
   • SqliteSaver for local persistence
   • PostgresSaver for production
6. Implement human-in-the-loop
   • Interrupt before/after nodes
   • Resume execution with new state
   • Approval workflows

Templates

• templates/graph.py — Main graph definition (StateGraph, edges, compilation)
• templates/state.py — State schema (TypedDict, MessagesState)
• templates/nodes/llm_nodes.py — LLM call nodes
• templates/nodes/tools.py — Tool definitions and ToolNode

Resources

• references/state-management.md — TypedDict, reducers, MessagesState patterns
• references/persistence.md — Checkpoints (InMemory, SQLite, Postgres)
• references/patterns.md — Agent, Supervisor, Orchestrator-Worker, Evaluator-Optimizer

Examples

• examples/agent-with-tools.py — Complete agent with tools, routing, and persistence

Use When

The user wants to build a stateful, multi-step agent workflow using LangGraph with nodes, edges, tool use, and optional human-in-the-loop or persistence.