activities.py

"""Basic LLM activity with LangSmith tracing."""

from dataclasses import dataclass

from langsmith import traceable
from langsmith.wrappers import wrap_openai
from openai import AsyncOpenAI
from temporalio import activity


@dataclass
class OpenAIRequest:
    model: str
    input: str
    instructions: str = "You are a helpful assistant."


@traceable(name="Call OpenAI", run_type="llm")
@activity.defn
async def call_openai(request: OpenAIRequest) -> str:
    """Call OpenAI Responses API. Retries handled by Temporal, not the OpenAI client."""
    # wrap_openai patches the client so each API call (e.g. responses.create)
    # creates its own child span with model parameters and token usage.
    # max_retries=0 disables OpenAI's built-in retries — Temporal's activity
    # retry policy handles retries instead, with full visibility in the UI.
    client = wrap_openai(AsyncOpenAI(max_retries=0))
    response = await client.responses.create(
        model=request.model,
        instructions=request.instructions,
        input=request.input,
        timeout=30,
    )
    return response.output_text