app/chat.py

"""Chat orchestration: run the agent against a user turn, stream tokens, return cost."""

from typing import AsyncIterator

from agents import Agent, Runner, SQLiteSession
from openai.types.responses import ResponseTextDeltaEvent

from .cost import CostBreakdown, cost_from_usage


async def chat_turn(
    user_message: str,
    agent: Agent,
    session: SQLiteSession,
    max_turns: int = 8,
) -> AsyncIterator[tuple[str, CostBreakdown | None]]:
    """Run one user turn. Yields (accumulated_text, cost_or_None).

    Cost is None on every yield until the final one — the final yield contains
    the same accumulated text plus the run's cost breakdown.
    """
    result = Runner.run_streamed(
        agent,
        user_message,
        session=session,
        max_turns=max_turns,
    )

    text = ""
    async for event in result.stream_events():
        if event.type == "raw_response_event" and isinstance(
            event.data, ResponseTextDeltaEvent
        ):
            text += event.data.delta
            yield text, None

    cost = cost_from_usage(result.context_wrapper.usage)
    yield text, cost