Yes, an AI Helps Us Answer You

First, the honest part

When you write to Cellarion — a support ticket, a question about one of your wines, a GitHub issue, an email — the reply you get back was probably drafted with the help of an AI. Not always, but usually.

We're telling you this not because something went wrong, but because you might eventually notice — the style is consistent, the answers arrive quickly, and they tend to contain a suspicious amount of detail — and we'd much rather you hear it from us than wonder whether you'd been quietly handed off to a bot. You haven't been. But the truth deserves more than a yes/no, so here it is properly.

Cellarion is very nearly a one-person project

Behind the app there is one person, a codebase, and a shared wine registry that thousands of bottles point at. Everything here is free — no ads, nothing paywalled — and the server bills are partly covered by a handful of voluntary supporters, who get nothing extra for it except our gratitude — the rest comes out of one person's pocket. The one thing nobody can donate is time — and time is what the work takes: bug reports to investigate, wine records to research and correct, drink windows to curate, features to build, and messages to answer, in more than one language, from more than one time zone.

One person, by the way, is less fragile than it sounds. The entire codebase is open source under the AGPL, and Cellarion is built to be self-hosted. If that one person ever walked away, nothing vanishes: the code stays public, your data stays exportable, and anyone can keep running it — or fork it and carry on. A one-person project, yes; a single point of failure, no.

Without help, the arithmetic is brutal. Messages would sit for days. Registry fixes would wait behind support, or support would wait behind registry fixes. The polite fiction of "we'll get back to you shortly" would become an actual fiction.

With help, every message gets read, researched, and answered — usually within hours. That's the trade, and we think it's the right one. Using AI here isn't a way of caring less about your message. It's the reason your message gets the attention it deserves at all.

What the AI actually does — and why you might even prefer it

Here's the part most people don't expect: the AI doesn't just phrase replies. Before answering, it investigates.

When someone reports "my dashboard is blank", the assistant doesn't reach for a sympathy template. It looks at the actual saved dashboard, runs the actual queries those widgets would run, checks whether the data comes back sound, and only then drafts a reply — one that can honestly say "your widgets are safe, the fault is on our side, and here's what we found." When someone asks why a wine record looks wrong, the reply usually follows an actual check of that record, its history, and the source it was imported from.

That's the difference between a canned answer and an answer with receipts. Frankly, a human alone rarely has time to dig that deep for every single message. The AI does — so the replies you get are often more researched than pure-human support would be, not less.

The same is true behind the scenes: a lot of the registry curation you benefit from — duplicate merges, misplaced producers moved to their right fields, drink windows reviewed wine by wine — is done with heavy AI assistance, checked against real sources, with every change recorded.

What it is not

• It's not a wall between you and a human. A person directs the work, reads the conversations, makes the judgement calls, and stands behind every reply. When something needs an actual decision — a policy, a promise, an apology — that decision is human.
• It's not a brush-off machine. The point of the assistance is to give your message more investigation, not less. If a reply you get ever feels like it missed the point, say so — a human will look again.
• It's not marketing automation. We don't use AI to spam, upsell, or generate engagement. It answers the people who wrote to us, and that's it.

The fine print, because it matters

Your messages are processed to answer them — that's all. They aren't used to train AI models. The same privacy rules that cover the rest of your Cellarion data cover your support messages too; nothing about AI assistance changes what our privacy policy promises.

And if you'd simply rather not: say so in your message, and you'll get a fully hand-typed reply. It might take a little longer. That's the honest cost, and you're welcome to choose it.

Why we wrote this down

Because trust is the whole product. Cellarion asks you to keep years of your cellar's history with us, and that only works if we tell you how things actually operate — including the unglamorous parts. We'd rather be the app that told you plainly than the app you found out about.

If this raises questions, the support box is right there. An AI will probably help answer you. Now you know why — and you know who's behind it.