LangChain Integration

Use LangChain normally, but route model calls through the KoreShield proxy.

Runtime Contract

• Chat route: POST /v1/chat/completions
• Optional RAG pre-scan: POST /v1/rag/scan
• Auth: Bearer JWT, X-API-Key, or ks_access_token cookie

Python Example

import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="deepseek-chat",
    base_url="http://localhost:8000/v1",
    api_key="unused-by-koreshield-provider-client",
    default_headers={
        "Authorization": f"Bearer {os.environ['KORESHIELD_JWT']}"
    },
)

result = llm.invoke("Summarize this safely")
print(result.content)

RAG Pattern

1. Retrieve documents from your vector store.
2. Call /v1/rag/scan with user_query + documents.
3. Filter/block on unsafe result.
4. Send safe context to /v1/chat/completions.

Notes

If a third-party LangChain utility expects older scan-style SDK methods, adapt it to call the two server endpoints above.