VoiceRun — Enterprise voice, deployed securely.

This example demonstrates how to use OpenAI's AsyncOpenAI client to build an agent that answers user questions with GPT-4.1-mini and responds with audio.

Features

Uses OpenAI's AsyncOpenAI client for LLM integration
Retrieves variables from the context (added as environment variables)
Handles session start and user text input
Streams LLM responses to the user
Responds with audio using TextToSpeechEvent
Simple, production-ready async pattern

Code Example

from openai import AsyncOpenAI
from primfunctions.events import Event, StartEvent, TextEvent, TextToSpeechEvent
from primfunctions.context import Context
from primfunctions.utils.streaming import stream_sentences

openai_client = AsyncOpenAI(api_key=context.variables.get("OPENAI_API_KEY"))

async def handler(event: Event, context: Context):
    if isinstance(event, StartEvent):
        yield TextToSpeechEvent(
            text="Hello! Ask me anything and I'll answer using GPT-4.1-mini.",
            voice="nova"
        )

    if isinstance(event, TextEvent):
        user_message = event.data.get("text", "")

        response = await openai_client.chat.completions.create(
            model="gpt-4.1-mini",
            messages=[
                {"role": "system", "content": "You are a helpful assistant."},
                {"role": "user", "content": user_message}
            ],
            stream=True
        )

        for chunk in stream_sentences(response):
            if "content" in chunk:
                yield TextToSpeechEvent(
                    text=chunk["content"],
                    voice="nova"
                )

How It Works

OpenAI Client Initialization

The AsyncOpenAI client is created once at the top of the file for efficient reuse across requests.

Session Start

When a session begins (StartEvent), the agent greets the user and prompts them to ask a question.

Handling User Input

On each TextEvent, the user's message is sent to OpenAI's GPT-4.1-mini model using the async client. The LLM's response is extracted from the API result.

Audio Response

The agent replies to the user by yielding a TextToSpeechEvent with the LLM's answer, using the 'nova' voice.

This example shows a minimal, production-ready pattern for integrating OpenAI LLMs into your agent using async Python and VoiceRun events.

Documentation