Is this supported with langgraph

[thedevd]

I want to check if this can also be used for langgraph ?

[KeepALifeUS]

Yes, Guardrails can work with LangGraph! Here's how:

Integration Pattern

Guardrails operates at the LLM call level, so you can use it within LangGraph nodes:

from guardrails import Guard
from langgraph.graph import StateGraph

# Define your guard
guard = Guard.from_rail(...)

# Use in a LangGraph node
def validated_llm_node(state):
    # Guardrails validates the LLM output
    result = guard(
        llm_api=openai.chat.completions.create,
        messages=state["messages"]
    )
    return {"validated_output": result.validated_output}

# Build graph
graph = StateGraph(State)
graph.add_node("validated_llm", validated_llm_node)

State-Based Validation

For more complex flows, you can also validate state transitions:

def validate_state_transition(state):
    # Use Guardrails to validate state before proceeding
    if not guard.validate(state["proposed_action"]):
        return {"action": "rejected", "reason": guard.error}
    return {"action": state["proposed_action"]}

Key Points

1. Guardrails validates outputs, LangGraph handles orchestration
2. Use guards in nodes that make LLM calls
3. State can carry validation results for downstream nodes

The two frameworks complement each other rather than overlap.