Launched this week
Raychis
Identify your plants. Learn your garden. All offline.
9 followers
Identify your plants. Learn your garden. All offline.
9 followers
A gardening companion in your pocket. Identify what you grow, and the more you use Raychis, the better it knows your garden, tailoring what to do next to your plants, your region, and the season. Unlike the big plant apps: no account, no cloud, no subscription wall to open the app. It works offline in the garden or on a walk, and when it isn't sure, it tells you, instead of guessing. The entire identification model ships inside the app and runs on-device. No server calls, ever.
Hey Product Hunt. I'm Matt, the maker. Here's the moment that started this.
A while back I photographed a seedling in my garden. An app told me, with complete confidence, it was basil. I left it to grow. It turned out to be annual mercury, which is mildly toxic. Basil isn't. The app never hedged and never offered an alternative. It just said basil.
That stuck with me, and it's why I spent the last year building Raychis.
I've spent fifteen years in engineering, the last five leading SRE teams for production AI systems. I'm not a botanist, just someone who gardens and got tired of tools that were confidently wrong.
Raychis is a gardening companion that identifies what you grow and learns your garden over time, tailoring care to your plants, your region, and the season. Two things make it different:
It's honest. When it isn't sure, it tells you, instead of guessing.
It runs entirely on your phone. The whole model ships in the app. No account, no cloud, no server calls, no subscription wall to open it. It works offline in the garden or on a walk, and nothing leaves your device.
It learns your garden over time the more you use it. It is able to tell you what is going on, what to look out for, what tasks you should be doing, and what would be good to grow. It is more than a plant ID app. It is a gardening intelligence app that will act as a true garden companion.
Fitting real identification into a phone's memory budget was the hard part, and the interesting one. It meant composing smaller specialised models rather than leaning on one big model in the cloud.
I'd love your thoughts, especially on the honest-uncertainty question: would you rather an app admit when it isn't sure, or just hand you a confident answer? I'm here all day to answer anything.