app/api/inference/predict/route.ts

/**
 * Inference Proxy — /predict
 *
 * Lets a separately-hosted frontend (e.g. the Vercel deployment, which has
 * no Python inference process of its own) reach this Space's internal
 * FastAPI service. Gated on a shared secret since it bypasses the
 * Firebase-authenticated /api/predict route entirely — this is
 * backend-to-backend, not meant to be called from a browser.
 */

import { NextRequest } from 'next/server'
import { rateLimit, getRateLimitHeaders } from '@/lib/security/rateLimiter'
import { createSecureResponse, addSecurityHeaders } from '@/lib/security/headers'
import { getInferenceUrl } from '@/lib/server/inferenceUrl'

const INFERENCE_TIMEOUT_MS = 30_000

function isAuthorized(request: NextRequest): boolean {
  const expected = process.env.INFERENCE_PROXY_TOKEN
  if (!expected) return false
  return request.headers.get('x-inference-proxy-token') === expected
}

export async function POST(request: NextRequest) {
  const rateLimitResponse = rateLimit(request, '/api/inference/predict')
  if (rateLimitResponse) {
    return addSecurityHeaders(rateLimitResponse)
  }

  if (!isAuthorized(request)) {
    return createSecureResponse({ error: 'Unauthorized' }, 401)
  }

  let inferenceUrl: string
  try {
    inferenceUrl = getInferenceUrl()
  } catch (error) {
    console.error('Inference proxy: service misconfigured:', error)
    return createSecureResponse({ error: 'Prediction service is not configured.' }, 503)
  }

  const formData = await request.formData()

  let upstream: Response
  try {
    upstream = await fetch(`${inferenceUrl}/predict`, {
      method: 'POST',
      body: formData,
      signal: AbortSignal.timeout(INFERENCE_TIMEOUT_MS),
    })
  } catch (error) {
    console.error('Inference proxy: upstream unreachable:', error)
    return createSecureResponse(
      { error: 'Prediction service unavailable. Please try again shortly.' },
      503
    )
  }

  let result: any
  try {
    result = await upstream.json()
  } catch {
    console.error('Inference proxy: upstream returned non-JSON, status:', upstream.status)
    return createSecureResponse({ error: 'Prediction failed. Please try again later.' }, 500)
  }

  const response = createSecureResponse(result, upstream.status)
  const rateLimitHeaders = getRateLimitHeaders(request, '/api/inference/predict')
  Object.entries(rateLimitHeaders).forEach(([key, value]) => {
    response.headers.set(key, value)
  })
  return response
}