TopolearnEvery graph must be compiled before it runs, and invoke is the simplest way to run it.
A builder is only a description. Call graph = builder.compile() to turn it into a compiled graph you can run. Compiling checks the structure (for example, that edges point to nodes that exist) and is where you later attach extras such as a checkpointer for saving state.
Run it with invoke, passing the starting state:
result = graph.invoke({"text": "hello graph world"})invoke runs node after node until it reaches END, then returns the final state as a dict. You only need to provide the keys your first nodes read; the rest are filled as nodes run.
Other ways to run the same graph:
graph.stream(...)yields results step by step (covered later),await graph.ainvoke(...)andgraph.astream(...)are the async versions, for programs written with Python'sasync/awaitstyle, such as most modern web servers,invoke(..., version="v2")(LangGraph 1.1+) returns aGraphOutputobject with.valueand.interrupts, which is easier to type-check.
See what you built: graph.get_graph().draw_mermaid() returns a diagram description you can paste into any Mermaid viewer (Mermaid is a text format for drawing diagrams), and in a notebook draw_mermaid_png() renders an image.
A compiled graph is reusable and safe to call many times; build it once at startup, not per request.
Example
# continues the previous lesson's builder
graph = builder.compile()
result = graph.invoke({"text": "hello graph world"})
print(result)
# {'text': 'hello graph world', 'words': 3, 'shout': 'HELLO GRAPH WORLD'}
out = graph.invoke({"text": "hi"}, version="v2")
print(out.value["words"], out.interrupts) # 1 ()
print(graph.get_graph().draw_mermaid()) # diagram sourceRemember
- builder.compile() validates and returns a runnable graph
- invoke(input) runs to END and returns the final state
- ainvoke/astream are the async versions
- get_graph().draw_mermaid() visualizes the graph
Watch out for
- Rebuilding and compiling the graph on every request instead of once at startup.