Operator’s manual · The AI chapter
The AI in the app, on one page
Only a handful of buttons ask an AI model. Nothing else in the app does, nothing runs when a page opens, and nothing uses a credit in the background. Every one of them is listed below, from the same list the app charges from.
The buttons
| Button | Where | Cost |
|---|---|---|
| Yeast suggestions | In the planner, under Choose it. | 1 AI credit |
| Plan assessment | In the planner, under the strains you chose. | 1 AI credit |
| Fermentation protocol | On a vintage's page, with Generate the fermentation protocol. | 1 AI credit |
| Blend advice | On the blend bench, after you search for blends inside a target window. | 1 AI credit |
| Why a ferment is flagged | In Active fermentations, with Ask why in a flag's note. | 1 AI credit |
| Reading a vineyard photo | On a block's card in the vineyard, and on a vineyard photo in the journal. | 10 free a month, then 1 AI credit each |
| Reading a product label | In Supplies, when you add your own spray product. | Free (Fermata pays for it) |
| Reading a bench sheet | On a lot's page, when you record a lab panel. | Free (Fermata pays for it) |
| Reading a Weighmaster Certificate | In the vineyard, when you weigh in fruit. | Free (Fermata pays for it) |
| Reading a filed operations report | In Reports, when you enter a return filed outside Fermata. | Free (Fermata pays for it) |
| Asking the cellar book | In the Ask box at the top of every page. | 30 free a month, then 1 AI credit each |
What all of them have in common
- It's labeled. You get suggestions, not recommendations, and reasoning, not instructions.
- It's checked. Every claim is compared with the catalog and your records. A strain the catalog doesn't have, a material that isn't on your shelf, a timing the app doesn't know, and a sentence with a quantity in it are all removed, and the screen tells you what was removed.
- It does no math. Doses, weights, matches, ratios and sulfur figures all come from the app. The AI model is given them but isn't allowed to repeat any of them.
- It doesn't change anything on its own. Loading a suggestion, keeping a revision and saving a product are always your choice.
- It reads your own paperwork for free. The readers copy a document you already have into a form you'd otherwise type by hand, and Fermata pays for those.
What the app won’t do to you over credits
You aren't charged when an answer comes back empty.
Fermata checks every claim in a reply against its catalog and your records, and throws out anything it can't back up. If nothing usable is left, you didn't get an answer. The same goes for a refusal or a reply that was cut off. We still pay for that request. If our checking throws out something it shouldn't, that's our cost, since we're the ones who can fix it.
When you run out, you get no answer, never a worse one.
The app won't switch to a cheaper AI model, shorten the reply, or reuse an old answer. You'd have no way to tell weaker advice from good advice. Instead, it tells you it can't ask right now and when your included asks come back.
You see what's left before you press the button, not after.
The count is shown beside every button that uses a credit. You shouldn't find out you're out of credits by pressing a button and having it refuse. A surprise like that makes software feel unreliable when it has simply run out.
Nothing buys more credits for you.
There's no setting that refills your credits when they run out, and there won't be one. Credits are sold in packs rather than billed by use, so you decide how much to spend before you spend it, instead of finding out afterward.
Everything the app works out itself is included: the protocol, the schedule, the lab calculations, and every conversion and check. Asking an AI model is what costs credits, because each question is a real request to another company's service, and you only ask by pressing a button. The readers that copy your own paperwork into a form also ask an AI model, and Fermata pays for those.