Build with the same engines.
Two endpoints, one canonical schema. The Taresense API turns heterogeneous commerce data — checkout pages, paper receipts, screenshots, URLs — into structured JSON your stack can rely on.
Receipt extractor
Drop any receipt image. Get canonical line items, fees classified by sub-category (resort / convenience / wifi / cleaning …), taxes broken out, mandatory flags, and arithmetic-verified totals.
curl https://taresense.com/api/receipt/extract \
-H "Content-Type: application/json" \
-d '{
"image": "data:image/jpeg;base64,/9j/4AAQSkZJRg…",
"sourceUrl": "https://www.airbnb.com/rooms/12345"
}'Response:
{
"id": "26cc1cc2-04ac-450c-9109-6516acf5c8ae",
"extraction": {
"status": "success",
"confidence": 0.92,
"merchant": { "name": "Fairmont Maui", "domain": "fairmont.com", "location": "Maui, HI" },
"currency": "USD",
"total": 295.69,
"basePrice": 89,
"feesTotal": 127,
"taxTotal": 39.7,
"tipTotal": 0,
"lineItems": [{ "label": "Resort fee", "amount": 42, "type": "fee", "mandatory": true, … }],
"fees": [{ "label": "Resort fee", "amount": 42, "category": "resort", "mandatory": true }],
"taxes": [{ "label": "Resort tax", "amount": 18, "rate": null }],
"metadata": { "model": "claude-sonnet-4-6", "extractionMs": 1812 }
}
}Price truth graph
Query the 90-day price-truth graph for any URL ever scanned by any Taresense user. Percentile rank, trend regression, forecast, cohort discrimination signal — the institutional reference for "is this actually a good price?"
curl https://taresense.com/api/price-truth/<urlHash>
Response:
{
"urlHash": "2fab8e…",
"status": "ready",
"currentPrice": 395.9,
"stats": { "min": 358.26, "max": 405.88, "median": 380.38, "percentileRank": 73, "trendSlope": -0.04 },
"forecast": { "direction": "stable", "confidence": 0.4, "expectedRange": { "low": 391.4, "high": 399.3 } },
"discrimination": { "significant": true, "spread": 0.076, "recommendation": "Cleared cookies / incognito sees ~$30 less on average." },
"recommendation": { "action": "try_other_cohort", "reason": "..." }
}Pluggable model backbone
Receipt extraction is provider-agnostic. The default chain is together → anthropic → mock. Once our fine-tuned Llama-3.2-Vision is live, it slots into the same chain — identical canonical output, no code change for consumers.
# Env (Vercel or your infra) RECEIPT_PROVIDER_CHAIN=together,anthropic,mock TOGETHER_API_KEY=... ANTHROPIC_API_KEY=...
Coming when we open the API
Right now both endpoints are live and CORS-open for the Taresense web + extension surfaces. We’re shipping API-key auth + metered billing for third-party consumers in the next few weeks. Drop your email and we’ll send you a key when access opens.