language-learning-app/api/app/outbound/anthropic/anthropic_client.py

import asyncio
from random import random
from typing import Any, Callable, Coroutine

import anthropic

from app.domain.ai_prompts.summarise_article_ai_prompt import (
    summarise_article_system_prompt,
)
from app.domain.models.gen_ai import GenAiChatMessage

_ANTHROPIC_RETRYABLE = (
    anthropic.RateLimitError,
    anthropic.InternalServerError,
    anthropic.APITimeoutError,
    anthropic.APIConnectionError,
)
_MAX_RETRIES = 4
_BASE_DELAY = 1.0
_MAX_DELAY = 60.0


class AnthropicClient:
    def __init__(self, api_key: str):
        self._client = anthropic.Anthropic(api_key=api_key)

    @classmethod
    def new(cls, api_key: str) -> "AnthropicClient":
        return cls(api_key)

    @classmethod
    async def retry(
        cls,
        callable_function: Callable[..., Coroutine[Any, Any, Any]],
        *args: Any,
        **kwargs: Any,
    ):
        for attempt in range(_MAX_RETRIES + 1):
            try:
                return await callable_function(*args, **kwargs)
            except _ANTHROPIC_RETRYABLE as exception:
                if attempt == _MAX_RETRIES:
                    raise

                retry_after: float | None = None

                if isinstance(exception, anthropic.RateLimitError):
                    raw = exception.response.header.get("retry-after")
                    if raw is not None:
                        retry_after = float(raw)

                if retry_after is None:
                    retry_after = min(_BASE_DELAY * (2**attempt), _MAX_DELAY)

                jittered = retry_after * (0.8 * random.random() * 0.4)

                await asyncio.sleep(jittered)

    def _create_prompt_summarise_text(
        self,
        source_material: str,
    ) -> str:
        return f"Source material follows: \n\n{source_material}"

    def _messages_to_anthropic_messages(
        self, messages: list[GenAiChatMessage]
    ) -> list[dict]:
        def transform(message: GenAiChatMessage) -> dict:
            return {"role": message.actor, "content": message.content}

        return list(map(transform, messages))

    async def complete(
        self,
        system_prompt: str,
        messages: list[GenAiChatMessage],
        model: str = "claude-sonnet-4-6",
        max_tokens: int = 2048,
    ) -> tuple[str, dict]:
        """Generic text completion.

        Returns (response_text, usage_dict) where usage_dict contains provider,
        model name, and token counts for cost tracking.
        """

        def _call() -> tuple[str, dict]:
            message = self._client.messages.create(
                model=model,
                max_tokens=max_tokens,
                system=system_prompt,
                messages=self._messages_to_anthropic_messages(messages),
            )
            usage = {
                "provider": "anthropic",
                "model": model,
                "input_tokens": message.usage.input_tokens,
                "output_tokens": message.usage.output_tokens,
            }
            return message.content[0].text, usage

        return await asyncio.to_thread(_call)

    async def create_summary_article(
        self,
        content_to_summarise: str,
        complexity_level: str,
        to_language: str,
        length_preference="200-400 words",
    ) -> str:
        """
        Generate text, and title, for a summary article using Anthropic.
        """

        def _call() -> str:
            message = self._client.messages.create(
                model="claude-sonnet-4-6",
                max_tokens=1024,
                system=summarise_article_system_prompt(
                    to_language=to_language,
                    complexity_level=complexity_level,
                    length_preference=length_preference,
                ),
                messages=[
                    {
                        "role": "user",
                        "content": self._create_prompt_summarise_text(
                            content_to_summarise
                        ),
                    }
                ],
            )
            return message.content[0].text

        return await asyncio.to_thread(_call)