What is crema?

Crema is a tool for home espresso enthusiasts who own a smart machine controlled by a project called GaggiMate, with early support for another one called Gaggiuino. It automatically pulls data from your recent espresso shots, sends that data to Claude, and gets back suggestions for adjusting your grind, dose, and brewing profile, based on your specific grinder and beans rather than generic advice. The tool needs to run on a separate always-on computer on the same home network as the coffee machine, since the machine only exposes its data over the local network and not over the internet. A Raspberry Pi is the recommended device, but any small always-on Linux or Mac computer works, and it needs Python 3.13 or newer. It also requires a paid Anthropic API account, since Claude is what performs the actual review, and this costs a small amount of money per shot reviewed, roughly a dollar or two a month for typical home use. Once set up, crema downloads new shot data and turns it into diagnostic information such as channeling risk and pressure stability, then has Claude review it and give the shot a quality score along with suggestions. If you want to try an unfamiliar bag of coffee beans, it can suggest a starting grind, dose, and complete profile based on your most similar past shots. Any suggested change to your machine's profile is shown to you first as a draft, and you must explicitly approve it before it is written back to the machine, especially if it changes settings that determine when a shot stops. A built in trends page charts your shot scores over time so you can see whether your adjustments are working, and you can log how each shot tasted so future reviews take your preferences into account. There is also an optional community feature where users can share anonymized shot data to help build a shared dataset linking brewing advice to outcomes, and an optional integration that displays your latest shot's score on a small pixel display called a Tidbyt. Crema is meant to work alongside GaggiMate's own app, not replace it, adding an AI review layer on top of the machine's existing controls.