Routing and Branching

Decide, then act

Real apps need different handling for different inputs: billing questions to one prompt, technical questions to another, small talk to a cheap model. RunnableBranch takes a list of (condition, runnable) pairs and a default; the first condition that is true selects the path. Conditions can be plain Python or the result of an earlier classifier step (often a small, cheap LLM call that returns a label). For more complex flows with loops and shared state, use LangGraph. Always include a default branch and test unusual inputs.

Routing by keyword, run

I ran this offline in a Python virtual environment with langchain-core 1.6.6, langchain-text-splitters 1.1.2 and llama-index-core 0.14.25. No API key or network call is needed because a fake model or a toy embedding stands in for the real one. The first matching condition wins; anything unmatched goes to the default route.

from langchain_core.runnables import RunnableBranch, RunnableLambda

route = RunnableBranch(
    (lambda x: "refund" in x.lower(), RunnableLambda(lambda x: "billing team")),
    (lambda x: "crash" in x.lower(), RunnableLambda(lambda x: "engineering")),
    RunnableLambda(lambda x: "general support"),
)
for q in ["I want a refund", "App crash on start", "Where is your office?"]:
    print(q, "->", route.invoke(q))

Output:

I want a refund -> billing team
App crash on start -> engineering
Where is your office? -> general support

Quick check: What should a RunnableBranch always have?

• A trained model
• Exactly one condition
• A GPU
• A default branch for inputs that match no condition