LangGraph Workflow Visualizer

LangGraph is a library from LangChain for building stateful, multi-actor AI applications as directed graphs. Each node in the graph is a function or LLM call, edges define control flow, and state is shared across the entire graph. It enables complex agent behaviors like loops, branches, human-in-the-loop, and parallel execution that are difficult to express with linear chains.

LangChain provides linear chains of operations (input → step1 → step2 → output). LangGraph adds graph-based control flow where nodes can loop, branch based on state, run in parallel, and maintain persistent state across turns. LangGraph is designed for production agents that need sophisticated routing logic, while LangChain is better for simple sequential pipelines.

StateGraph is the core LangGraph class. It defines a graph where each node receives the current state and returns state updates. You define: (1) State schema — a TypedDict of fields. (2) Nodes — functions that read/write state. (3) Edges — connections between nodes, either direct or conditional. (4) Entry/exit points — where the graph starts and ends. Compiled graphs can be invoked, streamed, or run asynchronously.

Conditional edges dynamically route between nodes based on the current state. Instead of always going to the same next node, you provide a function that examines state and returns the name of the next node (or END). This enables branching logic — for example: after a tools node, if there are pending tool calls, go to the tool executor; otherwise, go to END.

Nodes are the computation units — they receive state, do work (LLM call, tool execution, data transformation), and return state updates. Edges define routing — which node runs next. Edges can be direct (always go to X) or conditional (examine state and choose X or Y). The START and END special nodes mark graph entry and exit points.