How Declair tracks receipt OCR costs in production

The solution

Declair routes Anthropic calls through proxy.spillr.io and tags every OCR request with X-Spillr-Feature: receipt-ocr. No changes to the OCR prompt logic — only the client base URL and two headers.

import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic({
  baseURL: 'https://proxy.spillr.io/anthropic',
  defaultHeaders: {
    'X-Spillr-Key':     process.env.SPILLR_KEY,
    'X-Spillr-Feature': 'receipt-ocr',
  },
});

// Declair sends the receipt image + extraction prompt as usual.
const res = await client.messages.create({
  model: 'claude-sonnet-4-6',
  max_tokens: 1024,
  messages: [{ role: 'user', content: ocrPrompt }],
});

• →Receipt images and extracted text are never stored by Spillr — only token counts and USD cost.
• →The Spillr dashboard shows daily OCR spend, model breakdown, and request volume.
• →Budget alerts can fire before OCR spend spikes — useful when testing new prompts or models.