import anthropic

# ── Cliente sync (FastAPI sync endpoints)
client = anthropic.Anthropic()  # key desde env

# ── Cliente async (FastAPI async endpoints)
async_client = anthropic.AsyncAnthropic()

# ── Con retries y timeout custom
client = anthropic.Anthropic(
    max_retries=3,       # 429/500 → backoff
    timeout=120.0,     # 2 min por request
)

# ── Override por request (sin mutar client global)
client.with_options(timeout=30.0).messages.create(...)

# ── Chat endpoint NutriBot (FastAPI + SSE)
async def chat_nutribot(pregunta: str, historial: list):
    historial.append({"role": "user", "content": pregunta})

    async with async_client.messages.stream(
        model="claude-opus-4-8",
        max_tokens=64000,       # streaming → sin timeout
        thinking={"type": "adaptive"},
        system=SYSTEM_NUTRIBOT,
        messages=historial,
    ) as stream:
        # Yield tokens en tiempo real a la UI
        async for text in stream.text_stream:
            yield text

        # Acumular mensaje final para historial
        msg = await stream.get_final_message()
        historial.append({
            "role": "assistant",
            "content": msg.content  # ← lista de bloques, no solo texto
        })

# ── System prompt estático NutriBot (≈ 2000 tokens, no cambia)
SYSTEM_NUTRIBOT = [
    {
        "type": "text",
        "text": """Eres NutriBot, asistente de nutrición clínica...
[2000 tokens de contexto: guías dietéticas, alergias comunes,
protocolos de la clínica, restricciones legales, formato de respuesta]""",
        "cache_control": {"type": "ephemeral"},  # ← cachear aquí
    }
]

response = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=16000,
    system=SYSTEM_NUTRIBOT,
    messages=[{"role": "user", "content": pregunta}],
)

# Verificar cache hit (debe ser > 0 desde la 2ª llamada)
usage = response.usage
print(f"Cache write: {usage.cache_creation_input_tokens}")
print(f"Cache read:  {usage.cache_read_input_tokens}")
# Si cache_read es 0 → revisar silent invalidators (timestamps en prompt)

from anthropic.types.message_create_params import MessageCreateParamsNonStreaming
from anthropic.types.messages.batch_create_params import Request
import time

# ── 1. Crear batch con los 500 pacientes
requests = [
    Request(
        custom_id=f"paciente-{p['id']}",
        params=MessageCreateParamsNonStreaming(
            model="claude-opus-4-8",
            max_tokens=4096,
            system=SYSTEM_NUTRIBOT,
            messages=[{"role": "user",
                        "content": build_prompt(p)}],
        )
    )
    for p in pacientes  # lista de 500 dicts de pacientes
]
batch = client.messages.batches.create(requests=requests)
print(f"Batch ID: {batch.id}")

# ── 2. Esperar resultados (polling cada 60s)
while True:
    batch = client.messages.batches.retrieve(batch.id)
    if batch.processing_status == "ended": break
    time.sleep(60)

# ── 3. Recoger resultados y actualizar BD
for result in client.messages.batches.results(batch.id):
    if result.result.type == "succeeded":
        plan = result.result.message.content[0].text
        update_plan_paciente(result.custom_id, plan)