LangChain

pip install director-ai[langchain]

graph LR
    Q["Query"] --> LLM["LLM Chain"]
    LLM --> GUARD["DirectorAIGuard"]
    GUARD --> NLI["NLI scoring"]
    GUARD --> KB["KB retrieval"]
    NLI --> CHK{approved?}
    KB --> CHK
    CHK -->|Yes| OUT["Output"]
    CHK -->|No| FALLBACK["Reject / Retrieval fallback"]

    style GUARD fill:#512da8,color:#fff
    style CHK fill:#ff8f00,color:#fff
    style OUT fill:#2e7d32,color:#fff
    style FALLBACK fill:#c62828,color:#fff

Usage

from director_ai.integrations.langchain import DirectorAIGuard

guard = DirectorAIGuard(
    facts={"capital": "Paris is the capital of France."},
    threshold=0.6,
    raise_on_fail=False,
)

# As a chain step
result = guard.invoke({
    "query": "What is the capital of France?",
    "response": "The capital of France is Berlin.",
})

print(result["approved"])  # False
print(result["score"])     # ~0.35

In a Chain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini")
chain = llm | guard
result = chain.invoke("What is the capital of France?")