app.py

from fastapi import FastAPI, Header
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import Optional, List
from contextlib import asynccontextmanager
import os

from auth import verify_api_key
from credits import check_and_deduct_credits
from acra import run_acra_pipeline
from classifier_inference import warm_up

@asynccontextmanager
async def lifespan(app: FastAPI):
    print("Loading ACRA classifier...")
    warm_up()
    print("Ready \u2713")
    yield

app = FastAPI(title="ACRA API",
              description="Adaptive Contextual Retrieval Architecture \u2014 NurricAI",
              version="1.0.0", lifespan=lifespan)

app.add_middleware(CORSMiddleware, allow_origins=["*"],
                   allow_methods=["*"], allow_headers=["*"])

class IngestRequest(BaseModel):
    texts: List[str]
    metadata: Optional[List[dict]] = None
    namespace: Optional[str] = "default"

class QueryRequest(BaseModel):
    query: str
    namespace: Optional[str] = "default"
    top_k: Optional[int] = 5
    rerank: Optional[bool] = True
    use_web: Optional[bool] = False
    llm_endpoint: Optional[str] = None
    llm_api_key: Optional[str] = None
    llm_model: Optional[str] = None

class QueryResponse(BaseModel):
    answer: str
    sources: List[dict]
    credits_used: int
    credits_remaining: int
    complexity: Optional[dict] = None
    retrieval_source: Optional[str] = None
    cost: Optional[dict] = None
    cost: Optional[dict] = None

@app.get("/")
def root(): return {"status": "ACRA API is live \U0001F680", "docs": "/docs"}

@app.get("/health")
def health(): return {"status": "ok"}

@app.post("/v1/ingest")
async def ingest(body: IngestRequest,
                 x_api_key: str = Header(..., alias="X-API-Key")):
    user      = await verify_api_key(x_api_key)
    cost      = max(1, len(body.texts) // 10)
    remaining = await check_and_deduct_credits(user["id"], cost)
    inserted  = await run_acra_pipeline(
        mode="ingest", texts=body.texts,
        metadata=body.metadata or [{} for _ in body.texts],
        namespace=body.namespace, user_id=user["id"])
    return {"status": "success", "chunks_indexed": inserted,
            "credits_used": cost, "credits_remaining": remaining}

@app.post("/v1/query", response_model=QueryResponse)
async def query(body: QueryRequest,
                x_api_key: str = Header(..., alias="X-API-Key")):
    user      = await verify_api_key(x_api_key)
    remaining = await check_and_deduct_credits(user["id"], 1)
    result    = await run_acra_pipeline(
        mode="query", query=body.query, namespace=body.namespace,
        top_k=body.top_k, rerank=body.rerank, user_id=user["id"],
        use_web=body.use_web)
    return QueryResponse(
        answer            = result["answer"],
        sources           = result["sources"],
        credits_used      = 1,
        credits_remaining = remaining,
        complexity        = result.get("complexity"),
        retrieval_source  = result.get("retrieval_source"),
        cost              = result.get("cost"),
    )

@app.get("/v1/usage")
async def usage(x_api_key: str = Header(..., alias="X-API-Key")):
    user = await verify_api_key(x_api_key)
    return {"plan": user["plan"],
            "credits_remaining": user["credits_remaining"],
            "credits_reset": user["credits_reset_at"]}