A wine intelligence API
built from the ground up

FastCork isn't a general-purpose LLM with a wine prompt. It's a model trained exclusively on wine data — labels, reviews, appellations, and producer records — and served at the edge for sub-second responses.

The model

Most wine AI products are general-purpose models prompted to answer wine questions. FastCork is different: the model was built from the ground up for wine recognition, recommendation, and pairing. It was never trained to write code, summarise news articles, or answer geography questions. Every parameter is tuned for wine.

The result is higher accuracy on wine-specific tasks — producer name extraction, appellation disambiguation, grape variety classification — and much lower hallucination rates than models that treat wine as a niche sub-topic of general knowledge.

Infrastructure

FastCork runs on Cerebras CS-3 wafer-scale chips — dedicated silicon built for AI inference, not GPU clusters shared across thousands of workloads. Cerebras hardware delivers consistent, low-latency responses because the entire model fits on a single wafer. No batching delays, no cold starts, no contention.

Inference nodes are deployed across 25 datacenters globally, routed via Cloudflare's edge network. Requests are served from the node nearest to the caller. Median end-to-end latency is under 100ms from model start to first token — before network round-trip.

Training data

The model was trained on authoritative wine publications, appellation body records, and producer data accumulated over decades. Sources include:

Trusted by

FastCork powers wine features at companies across hospitality, e-commerce, and consumer apps.

Start building

Self-serve signup. Your API key is ready in under a minute. $5 to start, no commitment.