Where we left off

Ten lessons in, you've been building graph shapes by reading a series of add_node/add_edge/add_conditional_edges calls and mentally picturing the resulting shape. That gets harder fast once a graph has a dozen nodes and several branches. LangGraph can print the shape it actually compiled, so you can check it matches what you meant to build, this becomes genuinely useful debugging once graphs get bigger, later in this course.

The code, piece by piece

print(app.get_graph().draw_mermaid())

app.get_graph() returns a representation of the compiled graph's structure, every node and every edge, including the conditional ones. .draw_mermaid() renders that structure as Mermaid diagram source, a small text-based diagram language, as plain text. It needs no extra dependencies and always works, so it's the method this lesson leans on. You can paste the printed output into any Mermaid renderer (many Markdown viewers, including GitHub, render mermaid code fences automatically) to see the actual picture.

app.get_graph().draw_mermaid_png()

If you want an actual image file instead of text you paste elsewhere, .draw_mermaid_png() returns PNG bytes directly, write them to a file with open("graph.png", "wb").write(png_bytes). This needs additional dependencies to render remotely and isn't guaranteed to work in every environment, so treat it as an optional aside, .draw_mermaid() is the one to reach for by default.

Reading the output against the code that produced it

We rebuild Lesson 7's think/act loop (call_model routing to either tools or ending, tools routing back to call_model) specifically because it has both a branch and a cycle, the two shapes hardest to picture just from reading add_edge calls in order. Look at the printed Mermaid source and confirm you can find: the START arrow into call_model, the branch out of call_model (one path to tools, one to END), and the edge from tools back to call_model that makes it a loop. Being able to cross-reference a printed diagram against the code that built it is the actual skill here, useful the moment a graph is too big to hold entirely in your head.

Checkpoint

  • app.get_graph(): returns the compiled graph's structure, nodes and edges, including conditional ones.
  • .draw_mermaid(): renders that structure as Mermaid diagram source text, no extra dependencies, paste it into a Mermaid renderer to see the picture.
  • .draw_mermaid_png(): an optional alternative that returns actual PNG image bytes, needs more setup, not the default choice here.
  • cross-checking: reading a printed diagram back against the add_node/add_edge calls that produced it is how you verify a graph's shape matches what you intended, especially once it's too big to picture from code alone.

If anything here still feels unclear, ask before moving to Lesson 12, the beginner tier's checkpoint project.