Discover cocktails you'll love, learn what you can make with what you have, and build a home bar that's truly yours.
Get early accessMixeur runs two complementary ML models that work together. A flavour model maps your taste preferences against each cocktail's multi-dimensional profile. A companion model finds users who rate like you and surfaces cocktails that flavour matching alone would miss.
The two signals blend adaptively — leaning on flavour science for new users, incorporating community intelligence as your rating history grows. Large language models power the underlying flavour analysis, ensuring every cocktail profile reflects real ingredient chemistry, not keyword guesswork.
Every cocktail in our catalog has a verified multi-dimensional flavour profile. We'd rather have 350 cocktails understood deeply than 10,000 catalogued superficially.
Every suggestion comes with a plain-language explanation — which flavour axes matched, which users with your taste loved it, and why. No black boxes.
When we suggest an ingredient to buy, it's because it unlocks cocktails you'd genuinely love — not because a retailer paid for placement. Recommendation quality is never sacrificed.
We use machine learning and LLMs to understand flavour chemistry, find taste companions, and bootstrap new users — not to optimise engagement metrics or build advertising profiles.
Your taste profile evolves as you explore. The system tracks how your preferences shift over time, turning ratings into genuine self-discovery — not just a recommendation input.
Taste profiles power your experience, nothing else. No behavioural tracking, no data brokering, no third-party advertising profiles.
"Recommendation quality is never sacrificed for commerce."