Pattern guide

LangChain Integration Pattern

Privocio does not ship a native LangChain plugin. Use the OpenAI-compatible client for transcription, then pass structured text into your LangChain agent chain. Agent output mode reduces tokens before the LLM step.

Voice → transcript → agent

Separate STT (Privocio) from your agent LLM provider. Keeps voice data off shared training pipelines while using familiar LangChain patterns.

from openai import OpenAI
from langchain_openai import ChatOpenAI

# Transcription client — Privocio as Whisper backend
stt = OpenAI(
    api_key="YOUR_PRIVOCIO_KEY",
    base_url="https://api.privocio.com/v1",
)

def transcribe_audio(path: str) -> str:
    with open(path, "rb") as f:
        result = stt.audio.transcriptions.create(
            model="whisper-1",
            file=f,
        )
    return result.text

# Agent LLM (separate provider) consumes cleaned transcript
llm = ChatOpenAI(model="gpt-4o")
transcript = transcribe_audio("user_voice.wav")
response = llm.invoke(f"User said: {transcript}")

Output modes (Agent)