src/tutor/services/llm/base.py

"""LLM client contract.

Anything that can turn a chat into text satisfies `LLMClient` — the app, the
exercise generators and the evals only ever depend on this Protocol, never on
a vendor SDK (same pattern for ASR / TTS / storage).
"""

from collections.abc import Sequence
from typing import Literal, Protocol, runtime_checkable

from pydantic import BaseModel

Role = Literal["system", "user", "assistant"]


class ChatMessage(BaseModel):
    role: Role
    content: str


class LLMResponse(BaseModel):
    text: str
    model: str
    input_tokens: int | None = None
    output_tokens: int | None = None


class LLMError(RuntimeError):
    """A completion failed (network, auth, rate limit, provider error...)."""


@runtime_checkable
class LLMClient(Protocol):
    async def complete(
        self,
        messages: Sequence[ChatMessage],
        *,
        temperature: float = 0.7,
        max_tokens: int = 1024,
    ) -> LLMResponse: ...