Built for teams that need to get this right.
Lots of tools claim to detect fake receipts. Here's what makes airegco different — and why that matters for production use.
Six things that matter in production.
Two independent fraud signals
Most receipt-fraud tools give you a single opaque score. We combine AI-generated-image forensics with an OCR arithmetic-consistency check (does subtotal + tax = total?) — either one can flag a receipt, and you see which signal tripped, not just a number.
Restaurant-receipt-specific
v1 is trained and tuned on restaurant-format receipts, not a generic all-document detector. Hotel, taxi, and retail receipt support is planned as a v2 expansion once this vertical proves out.
Under 1s per check
The image-forensics model runs on CPU with no GPU queue or cold start. OCR verification, when enabled, adds one additional call. You get a combined verdict in the same HTTP round-trip.
Privacy by design
Receipt images are processed in memory and discarded after each request — never stored. We store only a SHA-256 hash for abuse detection. If you enable OCR verification, the image is sent to a GDPR-compliant OCR provider for text extraction and is not stored there either.
One API key, one endpoint
POST a base64-encoded receipt to /check. That's the whole integration. No SDK required, no webhook setup for the basic case, no OAuth flow for server-to-server calls.
Credits never expire
Buy once, use whenever. No monthly seats, no renewal pressure, no SLA tier forcing you into an annual plan you don't need.
airegco vs other detection tools.
How the detection works.
When you POST a receipt, it runs through two independent checks. First, an image-forensics model analyzes spatial patterns, texture signatures, and statistical features that AI generators consistently leave behind, producing an ai_probability in [0, 1]. Second — if OCR verification is enabled — we extract the subtotal, tax, and total from the receipt and check that the math reconciles, returning a math_consistent boolean. Either signal can independently flag a receipt for review — you get both results back, not a single blended score, so you can see exactly why a receipt was flagged.