Purpose-built for e-commerce
Optimized for catalogs, search, and discovery. Not generic text.
Request Access
Request Access
Natively multimodal: image + text
Trained on trillions of tokens, including social media. Vectra understands what products are, how people use them, and how they search for them. It enables vibe shopping — mood, occasion, aesthetic, intent.
Most models stop at the product description or in-house interaction data. Vectra understands the social context too: use cases, trends, and intent.
Optimized for catalogs, search, and discovery. Not generic text.
One embedding space for product photos, titles, descriptions, and queries.
Trillions of text and image tokens, across categories, languages, and markets.
Posts, reviews, comments. Trends and intent product pages miss.
"Bridesmaid dress for a summer wedding in Hawaii" works because Vectra learned from people who actually wrote about it.
Add and subtract meaning across text and images.
Top match
Deterministic dimensions
Vectra is built on deterministic dimensions — every dimension has a stable, consistent meaning across all embeddings, text and image alike. Other models scatter meaning unpredictably, so vector math produces noise. That's why multimodal vector arithmetic is only possible with Vectra.
Runs small. Integrates fast. No GPU cluster, no retraining pipeline, no MLOps, no drama.