When you ask, GardenTrack already knows the date, what you planted and what you did last week

Common questions

I tapped Ask Rafe about this bed and got asked which plant I meant

Fixed on 6 September 2026. Tapping the button on a bed or a plant is how you tell Rafe what you are asking about, so being asked to name it again was the app forgetting where you were standing one message ago. He now answers about the place — and if your question could mean several plants on it, he answers for them together (the lettuce is closest, about a week; the spinach after that) rather than handing the question back. He will still ask for a photo when he genuinely needs to see something; that is different, and he now names the bed itself instead of asking which one.

Talk said my bed was in full sun. I never said that.

That was a bug, fixed on 6 September 2026. If you have not recorded a bed's sun or whether rain reaches it, Rafe used to sometimes fill the gap — once telling us a bed was in full sun with no cover, and a moment later that rain did not reach it because it was covered. Both invented. He now knows the difference between you have not said and you said no, and will tell you he has no note rather than guess. You can set both on the bed's own page.

⚠ That fix reached one screen and not the other for a few hours on the same day, and we would rather say so. Asking from a bed, Rafe said he had no note. Asking from a plant, he still filled the gap — in one test describing a photograph taken five days ago of a plant that has no photographs at all. It is the same answer from both screens now, and the two places that sentence gets written have been merged into one, so a future fix cannot reach only half of it again.

On a computer, the past conversations panel only dimmed part of the screen

Fixed on 6 September 2026. On a wide screen the panel dimmed only the middle column, left the rest of the window bright, and clicking out there did not close it. On a phone it always behaved correctly, which is why it went unnoticed — the chat fills a phone, so there was no bright edge to see. Nothing was ever lost or mis-saved by it; it was only the dimming and the click-to-close.

Does Rafe know how much sun a bed gets?

Yes, from 6 September 2026 — if you have told us. Each growing space has a sun setting on its own page: full sun, part sun or shade. The plant page has always used it to say whether a crop is in the right spot; now Rafe can see it too, so why isn't my tomato fruiting? can take a shaded bed into account instead of guessing. If you have not set it for a bed, Rafe says nothing about that bed's sun rather than assuming.

Why isn't my greenhouse bed on the watering list?

Because you told us rain does not reach it — or that it does — and from 6 September 2026 Rafe can see that when you ask. If a bed is under cover, rain falling on your garden never waters it, so it keeps its watering job on a day the open beds skip theirs. That has always been true of the list itself; what is new is that Rafe can explain it. If you have not answered the question for a bed, Rafe says nothing about rain there rather than guessing — and you can answer it on the bed's own page.

Does Rafe look at my garden records before answering?

Yes — every time, and that is deliberate. For a few hours on 6 September 2026 he only looked when he judged the question needed it, which made general questions faster. We took that back the same day: asked about a plant from last season he answered I do not have a record of your garden from last year while the notes were sitting in the record. Being told your garden is not remembered is worse than waiting a few seconds, so he looks every time again.

What does the + next to the box do?

It adds something to the chat, since 5 September 2026 — one menu for all of it. Tap + and pick: Camera to take a photo, Photos to pick some from your library, A plant or A growing space to attach one as the thing you are asking about. From 22 September 2026 it arrives as its own card, with its picture, above the box — and that same card stays on your message once you send it. Tap it there to open the plant’s details and change them without leaving the chat. Attaching a plant does not change your garden; it just tells Rafe which plant you mean, the way you would attach a file. If you have no plants or beds yet, the menu says so and offers to add one.

Why are Rafe's answers sometimes short and sometimes long?

Because it decides, as of 7 September 2026 — we do not tell it. Until then it was instructed to keep answers under eighty words, and, three hundred lines further down the same instructions, that there was no word budget at all. Both are gone.

What we still ask for is plain language — about an eighth-grade reading level, no jargon without the ordinary word beside it — because an answer has to be readable standing in a garden with dirt on your hands.

⚠ If an answer is longer or shorter than it should be, that is worth telling us. It is a judgment it is making, not a setting we tuned, so the useful report is the answer itself rather than a number.

Can Rafe fix a plant I already added?

Yes, from 8 September 2026 — and until then he could not, and said so. Asked to update a tree, he answered that you would have to do it from the plant's own page. That was true: he could add plants and beds and had nothing that could change one.

Now he can correct what you tell him. Where a plant is growing, when it went in, what kind it is, how many; a bed's size, its name, how much sun it gets, whether rain reaches it. Say it in a sentence and he offers the change for you to accept.

⚠ It only changes what you actually said. If you mention three things, three things move and nothing else does — including anything he worked out himself. If he guessed your tree is about three years old, that stays a guess and never becomes a record.

⚠ He asks first, every time. Adding something wrong is easy to spot and easy to delete. Changing something quietly overwrites what you wrote, so it is never silent — and the correction goes into that plant's own story, dated, saying the record was corrected.

Tell him a month and he will take it. Say you planted it "in April" and he records a real day in April rather than asking which part — and it is not the 1st. He looks at your own last frost date: if that falls in April, he puts the plant in the day after it, because nobody plants into a frost. If your frost is earlier or later than the month you named, he takes the middle of the month and leaves it at that. Your word wins over the frost either way — if you say April and your frost is in May, he does not move your plant to May.

⚠ The date then shows as "roughly" everywhere it appears, because the month was yours and the day was Rafe's. Tell him the actual day later and the "roughly" goes away. A season is still too wide — say "in the spring" and he will ask which month, because six weeks is too much to guess inside.

Can I search my past conversations?

Yes, since 5 September 2026, and it works with no signal. Tap Chats at the top right; a search field sits above the list. Type a word and the list narrows to the conversations whose name or messages contain it, each with a line of what matched. Clear the field and the whole list is back. The search reads only what is on your device, so it works offline.

Can I change a question I already asked?

Yes, since 5 September 2026. Under each of your own messages there is Edit. Tap it and the words go back into the box for you to change; send, and it is asked as a new turn at the end of the conversation. The first question and its answer stay as they were. Only the words come back: a photo or a plant is attached fresh.

Can I rename or delete a conversation?

Both, from the list, since 5 September 2026. Tap Chats at the top right. Rename turns the name into a field: type, then press Enter or tap away to keep it (Escape keeps the old name). A name you give sticks, even if you gave it before asking anything. Delete asks once, in the row, with Delete and Keep. Deleting removes the conversation from every device.

Can I ask for an answer again?

Yes, since 5 September 2026. Under each finished answer there is Try again. Tap it and the same question is asked once more, as a new turn at the end of the conversation, with the same photo or plant attached if there was one. The first answer stays exactly as it was; both are on record. It is out of reach while Rafe is still writing — including the pause before the first word appears, which was a gap until 10 September 2026. Tapping during that pause used to ask the question twice and you got two answers; now the turn counts as open from the moment you send it. An answer that comes back empty still gets its Try again — that is exactly when you want it.

I spent several messages explaining my garden and nothing got saved. Why?

A real gap, fixed on 10 September 2026. Every change Rafe makes has to point at the words you used to ask for it — that is what stops him recording things you never said. He was only looking at your most recent message, so after a long back-and-forth he could not find them, and instead of saying so he tidied your words into a neat summary. A summary is not your words, so the change was refused, and the explanation he gave you was unreadable.

He now quotes you from whichever message you actually said it in, so a correction worked out over several messages gets recorded.

⚠ One case still needs a single sentence. If one change depends on facts spread across three separate messages — the variety in one, the place in another, the date in a third — there is no single thing you said that asks for it. Rafe quotes the closest one, and may still ask you to say it in one go.

Rafe showed me a list of changes with a button. What is that?

Reversible changes happen as you speak and can be undone; removals ask first — from 22 September 2026. Every change Rafe makes has to point at the words you used to ask for it — that is what stops him recording things you never said. After a few messages of sorting something out, there is often no single sentence that asks for the whole thing, so instead of refusing he shows you the card.

Each line is one real change, named: which zone, which bed, which plant. If nothing on the card removes anything, it is done the moment the card appears — you will see it marked Done, with one Undo if you change your mind. From 2 October 2026 it takes back everything on the card: the plants, the journal entries and any bed the card made. If one thing cannot be taken back, the card says so instead of saying your garden is as it was.

Or keep talking. If a line is wrong, say so — Rafe will adjust and show you a new one.

⚠ A change Rafe cannot describe exactly does not appear. If two of your beds share a name he cannot tell which you meant, so that one is left off the card and he asks you instead.

Two more things go on the card since 15 September 2026, whatever else was said. Anything Rafe wants to remove — an empty bed, an empty zone — is always a line you press yourself, never something he does by himself, because nothing is left to undo once it is gone. And when he has to guess where something goes — you said “the front porch” and there is no zone called that, so he picked one — the line says where he would put it, so a wrong guess is a thing you read rather than a thing you find later. On 11 September a gardener's bed was removed on a sentence about a different bed, and two new pots went into a zone they never mentioned; both are now lines on a card.

What happens if I ask with no signal?

Since 5 September 2026, it waits and sends itself. While your phone has no signal a line above the box says so. Ask anyway: the question is kept, and the moment signal is back it is sent once, by itself. A question the server refused for another reason is not re-sent; that one keeps its Try again.

Can I copy an answer?

Yes, since 5 September 2026. Under each of Rafe's answers there is a small Copy control. Tap it and the whole answer is on your clipboard; the label reads Copied for a moment and then comes back. It appears once the answer has finished arriving, and only under Rafe's answers, not your own messages.

Why did the page stop scrolling down while Rafe was still writing?

Because you scrolled up, since 5 September 2026. While an answer is arriving the page follows it as long as you are at the bottom. The moment you scroll up to read something, it stops following and leaves you where you are; the answer keeps growing below, and a small Jump to latest button appears above the box. Tap it, or scroll back down yourself, and the page follows again. Sending a message always takes you to the bottom.

Why do answers arrive in little jumps?

They no longer do, since 5 September 2026. An answer used to reach the screen in steps of about an eighth of a second, because each step was a write to the phone's own storage. Now the words are revealed one at a time as they arrive, with a small cursor at the growing edge, and the storage is written a little behind them. Longer answers keep their paragraph breaks again, too. Nothing about what is kept changed: reload mid-answer and you still have what had arrived; the whole answer is on record the moment it finishes.

Can I stop an answer while Rafe is still writing?

Yes, since 5 September 2026. While an answer is arriving, the Send button becomes Stop. Press it and it stops where it is: what had arrived stays as the answer, marked Stopped here, and you can ask the next thing straight away. Stopping does not hand the question back — the turn still counts against your allowance, because the work had already been done by the time you pressed it.

I told it about a plant it does not know — a durian, say. Is it recorded?

Yes, since 3 September 2026, and he says plainly when he has not. A plant our database does not hold is still a plant in your garden, so Rafe adds it under the name you used; it gets no watering, frost or harvest guidance, and the plant's page says so. Before this, Rafe sometimes said he would record such a plant and did not — the words and the record disagreed. Now a reply can only say a plant was recorded when it was, and if nothing was recorded he tells you and offers to add it anyway.

I told Rafe my tomatoes are Cherokee Purple and the plant came out called cherokee_purple. Why?

A slip between what you said and how it was stored — fixed 3 September 2026. Rafe was asked for the variety as a code word and wrote it down that way, and the app named the plant from the code. Now Rafe keeps the variety as you said it, a code that still slips through is turned back into words, and the plant is named Cherokee Purple. Plants already named that way can be renamed on their own page.

I tapped Rafe's button to add a plant and the variety does not link to anything. Why?

A near-miss of the slip above, on a different button — fixed 4 September 2026. When Rafe offers to add a plant and does not write it, you get a button to add it yourself. That button took the variety in your words and filed it in the slot meant for our catalog's own names, so the plant looked like it had a variety on it and nothing could use it — not its variety page, not the growing advice, not finding everything you grow of that kind. The plant page had already been fixed to look the name up first; this button had not, and neither had the one that adds another of the same plant. Both now look it up. If we hold that variety it links up properly, and if we do not, it is kept as your own wording rather than filed as a name we recognize. Plants added before this keep whatever was stored — opening the plant and picking the variety puts it right.

Rafe calls my radishes a "Radish row" and never asks which radish. Why?

Because for radishes, carrots, spinach, onions, potatoes, chives and crops like them, "which one?" is the wrong question — since 3 September 2026. They are sown by the dozen, in rows or bunches, or grow underground with only the greens showing; nobody points at one. So Rafe treats a sowing as the thing: one row of radishes is never a question, and if you sowed two rows on different days he asks by date — "the sowing from August 14 or the one from August 17?" — and your answer picks it. A tomato, a squash or a tree is still a plant you point at, and Rafe asks about those the way he always has.

Rafe asked which of two plants I meant. I answered with the date he gave me, and he asked again. Why?

That was a real gap, closed on 3 September 2026. Two radishes in one bed, sown three days apart: Rafe asked "the one from August 14 or the one from August 17?", you said "the one from August 17", and he asked again, because the part that finds a plant by name only ever looked at the name and the bed. Now your answer picks the plant: a date ("August 17", "Aug 17", "the 17th"), a word about which came first ("the earlier one", "the one I planted most recently"), or its place in the bed once you have put them in order ("the second from the left"). Words that name none of those leave the question open, and Rafe asks once more with the dates and places he can tell them apart by.

I have two rows of the same thing in one bed. It says it cannot tell which I mean and gives me nothing.

Fixed 1 September 2026 — it now asks you which sowing.

The app tells your plants apart by what they are called and which bed they are in. ⚠ Sow the same crop twice in one bed and both of those are identical, so it gave up: it would say it had not written anything down and offer you no way to fix it.

That is not a corner case. Sowing a row every two weeks is how you get lettuce all summer, and it made every one of those rows impossible to talk about.

Now it uses the date you sowed them — the one sown Aug 14, or the one sown Aug 17? — and you tap the one you meant. You typed those dates in, so you know which is which.

It still says nothing rather than guessing when two are genuinely identical — same crop, same bed, sown the same day. Two buttons you cannot tell apart are worse than being asked to say it again, because tapping one looks like it worked.

It told me I had not asked it to write anything down. I definitely did.

You did, and the message was wrong. Fixed 1 September 2026.

When the app cannot write something down it tells you underneath the answer, so a cheerful reply never stands over an empty record. ⚠ Two of those messages described what YOU did, and both of them appear when OUR side loses your words on the way through.

One said you had not asked it to write anything down. Another said what you mentioned was not something you said. Neither was about you — they happen when the part that reads your message does not carry your words forward properly.

They now say what actually happened: it did not catch it. Same information, same line, and it no longer tells you something untrue about yourself.

If you see it, saying the same thing again usually works — it is not a sign the app has stopped listening to you.

What does Rafe say to me the first time I open the chat?

From 9 September 2026, he reads your garden back to you. Your frost dates and where they came from, the plant you added and the spot it is in, and what you said you most want from your garden — the same facts the last setup screen showed you, said back so you can see they landed.

Then one question: when did that plant go in? Setup does not ask it, and it is the single date everything else is counted from — how far along the plant is, when its jobs come up, how long ago anything happened. ⚠ If you already told us the date, there is no question at all.

⚠⚠ He will not invent a job for you. Nik’s rule for the first run is that a wrong first task on day one is worse than no task, so the first message tells you what is known and asks one thing. Nothing else.

⚠ And it is the app talking, not a guess. Every line comes from something you entered, so it cannot get your dates or your plant wrong — if a line looks wrong, the record behind it is wrong, and fixing it on the plant’s own page fixes the message.

I listed two plants in one message and only the second one was added.

Fixed on 9 September 2026. "I put six tomatoes and three peppers in the back bed on Saturday" added the peppers and not the tomatoes, under an answer saying he had recorded both.

⚠⚠ It was the first thing you listed that went missing, every time, and that was not luck. Before Rafe writes anything he has to quote your own words back, unbroken — the rule that stops him writing down things you never said. In a list like that one, the bed and the date come at the end and belong to both plants, so only the LAST thing you named had an unbroken quote. Everything before it had the other items sitting in the middle of the words it needed.

Now two entries from one sentence can share its ending. If one quote picks up where the other leaves off and both finish on the same words, they are both your sentence. Everything else is unchanged — a word Rafe added, a word he skipped, or your words in a different order are all still refused.

Rafe told me my plant was fine and nothing was written down.

Fixed on 9 September 2026. "My coneflowers have gone brown and crispy on top — is something wrong?" got a good answer — those are seed heads, it is what the plant does in August — and left no trace at all. Ask the same thing next August and the app has no memory of ever having answered it.

⚠⚠ Being told nothing is wrong is about half of every real answer, so it was the answer you meet most often that the app could never remember for you.

Now a Write it down button appears under an answer that rules something out, with the thing you were worried about on it in the answer’s own words. Tapping it puts a dated line in that plant’s story. Nothing is written unless you tap.

⚠ Only reassurance works this way. A pest, a disease or a shortage still has to be something Rafe noticed and said he noticed — those go into a plant’s permanent story, and the app will not put one there by reading back over its own words.

I used Talk to remind me about something and no reminder came.

Fixed on 8 September 2026. "Remind me to net the cherries in June and prune the apples in December" got you "Done. You’ll get reminders in June and December" — and two notes on your trees dated that day, which is not a reminder and is not what you asked for.

⚠⚠ Rafe has no way to put a job on your list by himself, and he was using the only tool he had that writes anything. He even typed the word Reminder into the note.

Now a job you want doing later comes back as a card you tap — Put this on my list — with your own words on it and the month you said. Tapping it is what makes the reminder real, so nothing is on your list until you say so.

⚠ And if he ever claims a reminder is set when none is, the app says so underneath, and tells you where your words actually went.

Rafe said he moved my tree and it did not move.

Fixed on 8 September 2026, and until then he could not move a plant at all. Told "the apricot is not in the front porch, it is in the backyard", Rafe answered "your apricot is now marked as in the ground in your backyard where it belongs" — and the tree stayed on the porch.

⚠⚠ That is worse than nothing happening: the record was wrong and you had been told it was right. He had no way to move a plant. What he changed was what the roots are in — a pot, or the ground — which is not the same as which bed the plant stands in, and he made a garden zone called backyard that nothing could go into.

Now he can move a plant, and a move is a real move: the plant leaves the old space, its place in the old bed's drawing closes, and its own story gets a line saying it moved and where to. You approve the change first, as with any correction.

⚠ A place is not a bed, and he will not pretend otherwise. "The backyard" is somewhere your beds are; a plant lives in one of them. Say a place with no bed in it and Rafe will not guess — the app asks which space you mean, and offers to make one there.

I used Talk to move everything out of a bed, and Rafe offered to move the bed.

Fixed on 23 September 2026. Told "move everything in the orchard to the side yard bed and delete the orchard", Rafe offered to move the orchard itself, trees and all, and then to take out the backyard it had left empty. You meant the trees.

Now every plant gets a move of its own into the bed you named, and the orchard is taken out last, once it is empty.

I said "the back bed" and it asked me which bed I meant.

Fixed on 8 September 2026. Rafe knew which bed. The part that checks the answer was comparing your words to the bed's name letter for letter, so "the back bed" did not match a bed called Back bed and the question came back to you — about the bed you had just named.

⚠ It only went wrong when Rafe named the bed. If he left that out, an older, looser check read the same words out of your sentence and the plant landed. So supplying the answer was worse than leaving it blank, which is the wrong way round.

Now your words are checked against your own list of beds. A bed whose name sits inside what you said is that bed. If two of your beds could both be what you meant, it still asks — and tells them apart by what is growing in each, because two buttons you cannot tell apart are worse than the question.

I told Rafe the day I planted something and the plant has no date on it.

Fixed on 8 September 2026. "I put six tomatoes and three peppers in the back bed on Saturday" added both plants, with the right numbers, and threw Saturday away without saying so.

Two names for one thing, and ours. Rafe has one word for a planting date when he is describing what you said, and the part that stores the plant expected a different word. Nothing rejected the mismatch, so the plants were created and the date went nowhere.

⚠ A plant with no date is not a small thing. Every reminder, every stage, every "three weeks ago" in the app is counted from the day it went in. A plant without one quietly stops producing jobs.

The two vocabularies are now read as one, for the date, the number and the bed alike — and there is a test that drives that exact sentence and reads the saved row back.

I said I planted something and it told me it had not written it down. Why not just ask me where?

It does now, from 1 September 2026.

If you say you planted something without saying which bed, the app will not guess. ⚠ That part is deliberate and is not changing — a plant filed in the wrong bed is worse than one it asked about, because everything after that is about the wrong place.

What was missing is that it did not ask either. It said it had not written anything down and left you to start again. Now it lists your beds and you tap one.

It only asks when the answer is a real choice. If you have one bed there is nothing to ask. And if two of your beds have the same name it says nothing rather than showing you two buttons you cannot tell apart — tapping one of those looks like it worked, which is worse than being told plainly that nothing was saved.

Naming the bed in your message still works and is still fastest: “three basil in Herb Corner” is written down without any question at all.

I told Rafe the plant, then the kind, then which bed — three messages — and he said he had not added anything.

Fixed on 3 September 2026, and the cause was punctuation.

Before Rafe adds a plant he has to quote your own words back. That is the rule that stops him writing down things you never said, and it is not going anywhere. He had already been taught that an answer to his own question still counts as your words — so tomatoes in your first message and the back bed two messages later were both fair to quote.

What went wrong is that he quoted all three messages as one sentence, with full stops he put in himself. We were checking that quote letter for letter. Those full stops were never in anything you typed, so the check decided the words were not yours, and nothing was added — to a gardener who had just answered every question he asked.

Now the check reads the words, not the punctuation. The same words in the same order, and it still refuses anything you did not say: a word Rafe added, a word he skipped over, or your own words shuffled into a different order. Spacing, capitals, commas and full stops are no longer evidence of anything, because they never were.

My answer stopped in the middle of a word. What happened?

It ran into the length limit, and from 1 September 2026 the app says so.

Every answer has a maximum length. A long one occasionally reaches it and stops wherever it happens to be — sometimes mid-sentence, sometimes mid-word. We knew when that happened and said nothing, so a reply ending in the middle of a word looked like your phone had lost signal. Now there is one short line underneath: that answer was cut short.

It does not tell you to ask again for the rest, and that is deliberate. Asking again starts a brand new answer rather than picking up where the last one stopped — so “ask for the rest” would be promising something the app cannot do. We would rather say nothing than that.

What does work is asking something narrower. A question about one plant, or one problem, leaves plenty of room for a full answer. If you hit this often, tell us — the limit is a number we chose and we are watching how often it bites.

And from 2 September 2026 the limit itself is twice what it was. It was sized for words alone, and an answer about a photograph spends part of it looking before it writes — so the words were what got cut. The reply that stopped at “staying on those older le” had used the whole limit; a reply that long now has room to finish, and the short line still appears on the rare one that does not.

I told Rafe I planted something and he said "got it" — but the plant is not there.

That could happen until 1 September 2026, and it is fixed. Tell Rafe you just planted something in a particular bed and he writes the plant down for you. Some of the time, though, he would reply “got it, three basil seedlings going into Herb Corner today” — and save nothing at all.

The reason was hidden and it was ours. The part of Rafe that writes the reply had worked out which bed you meant. The part that saves the plant was only allowed to look in one place, and that place was sometimes empty. So one half knew and the other half could not see it.

Now it reads the bed from what you actually said. Name one bed and the plant goes in that bed. Name no bed at all and it still holds off rather than guessing, which is the behavior we want to keep — a plant filed in the wrong bed is worse than one he asked about.

We found this by using the app, not by reading the code. The same sentence, sent seven times, worked five times. Every test we had passed, because every test handed the code the one thing the live assistant kept forgetting to pass.

How do I know it actually read my notes and did not just make something up?

You cannot see the list any more — removed on 22 September 2026 — but Rafe still only answers from your notes when he has them.

From 1 September to 22 September the answer was followed by It used what you had written down, naming the entries he leaned on. We cut it: it was one more thing to read under every reply, and it never changed what to do next.

Nothing about how Rafe reads your garden changed. When your notes answer the question, he still leans on them; when they do not, he still falls back to what we know about the crop in general. There is just no readout of which underneath any more.

This is the thing a general chatbot still cannot do. It does not have your notes, shown or not.

It used to list which of my notes an answer was based on. Sometimes there is nothing there now.

That list is only ever entries we actually handed the assistant, and from 26 August 2026 we check that rather than trusting it.

When he answers, he marks the journal entries he drew on, and the app shows them under It used what you had written down. ⚠ We were not checking those marks against what we had sent. So an entry could appear on that list when it had played no part in the answer at all — most likely one he had seen earlier in the same conversation and remembered, rather than anything he made up.

A list you cannot trust is worse than no list. The whole point of it is to let you check the answer against your own record, and that only works if every line on it is genuinely what was read.

So sometimes you will now see fewer entries, or none. ⚠ None does not mean the answer ignored your garden — a great deal of what we send is your plants, your dates and your beds rather than individual notes, and only notes can be listed this way. It means nothing on that list would have been honest.

Nothing about the answer itself changed. The same information goes to the assistant as before; only the list underneath is stricter.

Rafe asked what kind of tomato it is, days after I added it. Why now?

Because you were talking about that plant, and its kind is not on record — new on 4 September 2026.

When you add a plant without saying which kind it is, Rafe asks in that conversation. He did not ask again after that, so a plant could sit for weeks with no kind and nothing would ever raise it. Now, when you mention that plant on a later day, he asks once — one plain question at the end of his answer — and the same card appears under it that you get after adding a plant: say the kind, or have it made into a task.

Only about the plant you are talking about, never twice in one conversation, and not at all if you already asked for a task about it. If you do not know, the task is the honest answer, and the plant keeps the jobs it has.

Can Rafe remember things about me, not just about the garden?

Yes, from 4 September 2026 — and only what you tell him to.

The garden is already his memory for anything the garden can hold: a note you wrote about aphids in one conversation reaches the next one without being asked. What a garden cannot hold is a fact about you — I never spray, the back bed floods, I travel every August, you keep telling me to plant out too early for my spot. Say remember that… in the chat and he writes it down as one line, in your words, with the sentence it came from under it.

The list is on your account page under How you garden, with a Delete on every line, and you can add a line there too. He reads it on every answer and follows it over his own advice.

He never adds a line on his own. Other assistants also guess facts about you from your chats; Rafe does not, because a guess nobody said, steering every answer, looks exactly like a right one. And a line you delete — or tell him to forget — is gone from his very next message. There is no store in between for it to linger in.

What is the line above the answer while I wait?

The step Rafe is on — and from 22 September 2026 it starts the moment you send, not once the first step lands.

The instant you send, the line says Looking at your … and names whatever you attached — a plant, a bed, a zone — or Thinking… when you attached nothing. As he works it becomes whatever step he is actually on: Reading back through your journal…, Searching your whole journal…, Looking at your photo…, Checking your beds and where things are…. It changes as it moves from one step to the next and disappears the moment his answer starts.

It used to go quiet until the first step landed, and a step he only takes after writing his reply — noting what a photo showed — could pop in above an answer that was already finished. Now something true is always there from the moment you send, and a step taken after the words have already started never appears above them.

Is there a limit on how much I can ask? Profile used to say there was none.

Yes, from 23 September 2026: 75 credits a week on a paid plan, the beta included. The free plan works differently. It comes with 5 one-time Talk credits, about two dozen messages, and they don't reset.

Until then an open plan had no weekly limit, and Profile said so. When we softened that line on 24 August 2026 we promised the no-limit part would not be quietly withdrawn, so this is us saying it plainly. Profile now tells you what you get: 75 credits every Monday, and anything you do not spend carries over, up to 300 in all. Tap the ring beside the mic to see what is left this week.

Behind the credits there is still a safety ceiling on every account, so that a bug in our code, or somebody else getting into your account, cannot run up an enormous bill overnight. Credits run out long before anyone could reach it. If you ever do meet it, get in touch and we will sort it out — it would mean something unusual happened, and we would want to know.

A question failed and said something broke on your end. What does that mean?

It means the fault was ours rather than the answering service’s, and we would rather say which.

Three things happen when you ask a question. We count it against your account, we look up whatever the question needs from your own records, and then we ask the model. Only the last one is the answering service.

⚠ Until 27 August 2026 all three failed with the same sentence — “could not reach the answering service” — so if the first two went wrong we were telling you the model was down while it was answering everyone else perfectly well.

Now it says whose fault it was. If it was ours, it says so plainly. Either way, trying again is worth it — most of these last seconds rather than minutes, and nothing you typed is lost.

⚠ If it keeps happening, tell us. A message about our end means something on our side needs looking at, and we would rather hear it from you than find it later.

It rained hard and the app still asked me to water.

Fixed 1 September 2026. A good soaking has counted as a watering since August — but it was clearing the wrong list.

There are two names for watering inside the app and six for the checks you actually get shown: water seedbed, water tray, water bed, and so on. ⚠ The rain was only clearing the two you never see, so it rained, the app knew, and it asked anyway. It clears all of them now.

And it tells you when the rain did it. Where the check comes back — a seedbed is checked every ten hours, so yesterday's rain does not cover today — the job now carries the reason with it: Rain watered this yesterday (0.83 in) — seedbeds dry fast, so it's due again. ⚠ Both of those are true at once, and a job that reappears with no explanation reads as the app forgetting.

How much counts has not changed: roughly a third of an inch, added up across three days, because drizzle on three days soaks the same ground as one shower.

⚠⚠ A plant under glass or in a seed tray indoors is credited nothing, however hard it rained — none of it lands on them, and telling you a greenhouse tomato had been watered by a storm it never felt is the kind of wrong answer that costs a plant.

I opened a bed and asked what was wrong, and Rafe asked me which plant I meant.

We told him not to do that on 6 September 2026, and later the same day we found it still sometimes did.

Opening a bed and asking about it IS you telling him which one, so asking you again is the app forgetting where you were standing. The instruction not to was already there — we tested 147 conversations and he still did it in about one answer in sixteen, nearly always on a bed holding several different plants when the question was about a symptom.

⚠ We had reported it fixed on a much smaller test. Twenty-four conversations came back clean, which at that rate happens about one time in five by luck. That was our mistake in how we checked, not a change in the app.

What changed now: rather than repeating the rule, he is shown the shape of a good answer at the moment your bed is in front of him — name what is on it and how long it has been in, and answer for the bed. He also had a habit of replying “send me a photo” without naming anything on the bed. Asking for a photo is often exactly right; asking instead of answering is not.

Later the same day we found one more gap and closed it. The change worked for questions about timing and not for questions about symptoms — ask what is wrong with it from a bed with three crops and he would still ask which plant, more often than before. The reason was the example we gave him: it was about planting dates, so that is the shape of question he learned. He now gets the symptom case too, and answers it by naming the plants the symptom could belong to and what it would mean on each.

If he still asks you which plant, tell him in the chat — that reaches us.

Can I ask about a job on my Today list?

Yes, from 6 September 2026. Open the job and there is an Ask Rafe about this button under the headline. It takes you to the chat with the job attached as its own card above the box, so you can ask why it is due, whether it can wait until the weekend, or how to actually do it — without typing out which job you mean. That card stays on your message once you send it (from 22 September 2026).

The top of the chat shows the job you came from, and tapping it takes you back.

The buttons under the job are unchanged. Those are what you DO with it — done, not today, put it in the diary. This one is what you ask ABOUT it, which is why it sits above them rather than among them.

⚠ One thing worth knowing, because it explains a difference you might notice. A plant and a bed are things we have saved. A job on your list is not — it is worked out fresh on your phone every day from what you are growing and what the weather is doing. So when you ask about a job, what travels is a description of it in words rather than a reference to a saved record, and that description is what is kept on your conversation. It means the note still makes sense next season, when the job itself is long gone.

Can I ask about a whole part of my garden, not just one bed?

Yes — from 9 September 2026 a garden zone has one too, so every place in your garden can be asked about.

Open the zone — the level above your beds, the backyard that holds the orchard and the raised bed — and tap Ask Rafe about this part of the garden. It goes to the chat with the zone attached as its own card above the box, and you can switch it off if you meant something else. From 22 September 2026 that card stays on your message once you send it.

Rafe is told what is in it: each growing space in that zone, how many plants are in each, and what they are. Not every plant in detail — a zone can hold six beds, and the detail is one tap further in. If he needs more he looks it up.

It is a different question from asking about one bed. How is the backyard doing is about the place as a whole; tapping a bed inside it asks about that bed.

And there is a way back, the same as everywhere else — the top of the chat shows the zone you came from.

Can I ask about a growing space, not just a plant?

Yes — from 5 September 2026 a growing space has the same button a plant has.

Open the bed, pot or tray and tap Ask Rafe about this space. It takes you straight to the chat with that space attached as its own card, with its picture, above the box, exactly as a plant is, and you can switch it off if you meant something else. From 22 September 2026 that card stays on your message once you send it.

And there is a way back. From 6 September 2026 the top of the chat shows the bed or plant you came from — ← Back bed — and tapping it returns you to that page. It stays there while you talk, so you can ask, read the answer, and go back to what you were doing. Until then the button opened a small panel over the page instead, which took you to the chat when you sent anyway, so it was a stop on the way rather than a place.

Nothing is sent until you send it. The button opens the box with the space attached; the words are still yours to type or speak. And a plant and a place can travel together — ask about your tomatoes on the front porch and both go with the question.

The plant version has been there since September and works the same way.

I asked about a plant and sent a photo. Where did the photo go?

Onto that plant’s page — and from 1 September 2026 the app tells you so.

If you tapped Ask Rafe about this plant from the plant itself, we take you at your word: you were standing in front of it, so the photo is filed there, dated the day you took it.

Under Rafe's reply you will now see a line like Photo kept on your Wax Bean, with the name linked. Tap the name to go straight to it.

⚠ If it picked the wrong plant, tap Not that plant? The photo comes back off that plant — the journal entry goes with it — and the app asks you which plant it was, the same way it does for any photo it cannot place.

Why it changed: it used to file the photo and say nothing. That was worse than it sounds, because a photo that has been filed no longer shows up under Photos with no plant — so if it guessed wrong, there was no way to notice and no way to fix it. Now it says what it did, and you can undo it in one tap.

Tapping the Talk tab at the bottom of the screen is different and files nothing. That is you moving around the app, not you pointing at a plant.

And from 2 September 2026 the filing reaches our side as well as your phone. Until then, if the upload finished before you said which plant — which it usually did — the copy on our side kept no plant for it, so Rafe could say he had no photograph of a plant you had three of, and a new phone got the photos without their plants. The plant now follows the photo the same way every other change does.

Can I keep an answer? It seems a shame it only lives in the chat.

Yes, from 1 September 2026. Under a reply about one of your plants there is a button: Save as a note on your Sungold. Tap it and the note opens for reading first — the exact words that will land on that plant’s page, in a box you can edit — then Save writes it or Cancel closes it with nothing written. From 22 September 2026 that is how it opens; before then the tap saved straight away.

Why it is worth doing: next August the same lower leaves will look just as alarming, and the note will be on the plant rather than buried in a conversation from last year. ⚠ That is the whole promise of the app — it remembers your garden so you do not have to — and until this, the best thing it ever told you was the one thing it forgot.

It saves the gist, not the whole reply — edited or not. A journal is a list you scan, and a page of text in one entry is not scannable; the full answer stays in the chat. It is marked read by the app, so a season later you can tell what we worked out from what you wrote yourself.

It only appears when we already know which plant you meant — if we do not, you get asked which plant instead, and answering that is what files it. Ignoring it costs nothing.

Who is talking to me on the Talk screen?

Rafe, and from 1 September 2026 the screen says so before you type anything.

Opening Talk with nothing in it used to give you a heading and a paragraph that both described the app from the outside rather than being his own words. The Talk screen opens with a one-line hello that changes with the time of day. From 22 September 2026 that is all a new chat shows.

Some answers are long. Are they meant to be?

Yes, when the question needs it. We looked at shortening them and decided not to — explaining something properly to somebody new to growing takes the words it takes, and a clipped answer that leaves out the reason is not actually kinder.

What we did instead was make a long one easier to read, on 1 September 2026. When an answer has sections, those sections now look like sections: a heading was set in exactly the same size, weight and color as bold text inside a sentence, so while you were scanning, nothing marked it out. It is a touch heavier now. Paragraphs had less than half a line between them and ran together into a block; there is a little more air.

⚠ Nothing shouts. A heading in a four-paragraph answer is a signpost, not a title page — headings that shout would turn a two-sentence reply into a page of admin. And we judged all of it at phone size, because that is where you read it.

Can a long conversation get too full? Other AI apps show a bar that fills up.

No, and not because we have not got round to building the bar — the app is put together so a conversation here cannot get too full for the assistant to handle.

Every message you send carries the conversation with it, so the assistant knows what you were talking about. We keep the whole thing, up to a size limit, rather than counting exchanges — a typical conversation is remembered in full, and only once one runs on for a good while does its earliest part start to fall away, oldest first. Each message is capped in length too. Photos ride on the message you attached them to and are described rather than re-sent afterwards.

Add it all up and even a long conversation in a big garden stays well inside what the assistant can hold — the size limit on the conversation is what keeps it there. It does not creep up on you, because it cannot.

⚠ So there is nothing useful for a bar to show. A gauge that can never move is not telling you anything, and the apps that show one need it because there you genuinely can run out. We would rather say the true thing on this page than draw a dial that only looks reassuring.

One practical note: a very long thread does forget the early part of itself. It takes dozens of back-and-forths to get there, and what goes is the oldest end. If a conversation has run on that long, ask your question freshly rather than saying “like I said before” — or just start a new chat, which costs you nothing.

How can I tell which journal entries the app worked out and which ones I logged myself?

By reading them — the entry says so, from 20 August 2026. An entry the app read rather than one you tapped ends with where it came from: "Aphids — from what you told the chat", or "Seen when the plant was added — from your photo". Anything without one of those is something you recorded yourself.

It is in the words rather than an icon on purpose. A little marker on every row clutters the thing you came to read, and it only helps on the screen it was built for — the note travels with the entry everywhere the entry goes.

Why it matters more than it sounds: the whole point of this app is that it remembers your garden so you do not have to. Reading back through October, "I watered this on Tuesday" and "a photo suggested the soil looked mulched and I agreed" are not the same kind of fact. They age differently and they are worth different amounts as evidence. You should never have to wonder which one you are looking at.

Nothing is ever written without you agreeing to it first, and that has always been true. This is about what you can tell later, not about what gets saved.

Where did the three suggested questions go?

Off the chat, on 22 September 2026 — a new chat is a greeting and the box, and nothing else. Each of the three has a screen that answers it better. This week's jobs are on Today. The yearly jobs you can still catch are under Catch up in the menu, and on the Calendar. Each plant's season is drawn on the Calendar. Ask in your own words and Rafe reads the same list Today shows.

I told the chat my carrots had their proper leaves. What happens to that?

It gets recorded, and any jobs that were waiting on it come straight back to your list. You do not need a button — say it in your own words and the reply tells you what changed: "Noted — thin seedlings is back on the carrot's list." If nothing was waiting on it, we record it and say nothing, because "noted" on a list that has not changed is the app taking credit for filing. If we are not confident about what you meant, we record nothing rather than guess.

One thing worth knowing, because it was wrong for the first hour this was live. A message like that did not carry which plant you meant — it was filed against whichever plant the conversation was already about. So mentioning your carrots in a thread about the tomato put the note on the tomato, and the app then offered jobs for leaves that plant had never grown. Nothing about the record looked wrong afterwards: right observation, right date, wrong plant. Now, if we cannot tell which plant you mean, we hold off instead of filing it somewhere plausible. Name the crop and it records exactly as before. This is the same rule we already use for pests — guessing which plant a problem belongs to is worse than asking.

I have a lot of plants. Does the chat see all of them?

Up to forty at a time, and from 26 August 2026 it tells you when there are more.

Every question you ask sends the chat a list of what is in your garden, and that list is capped so the answer stays fast. Under forty plants — which is most gardens — it sees everything and nothing here applies to you.

Over forty, what changed is which forty. It used to send the forty most recently planted, so the first thing dropped was whatever had been in the ground longest. An apple tree planted five years ago fell out before this spring’s lettuce did — and the answer would come back as though the tree did not exist, which reads as the app having forgotten it.

Now it takes twenty from each end: the twenty longest-established and the twenty most recently planted. What gets left out is the middle of the season, where one row of beans is much like another. It is not a judgment that trees matter more than lettuce — it is that cutting from one end always loses something you would have wanted.

And it now says when the list was cut, so the answer can tell you it might not be seeing a plant rather than quietly assuming that plant does not exist.

The same applies when you ask about one bed. If a single bed holds more than forty living plants, the chat lists the forty oldest and tells you how many are in there altogether — so the number you see is the number in the bed, not the number we sent.

I grow carrots in two beds. Does the chat know which one I mean?

If you tell it, yes. If you do not, it holds off rather than guessing — and that was wrong until 18 August 2026. In a conversation about your front bed, saying "the ones in the back have their leaves" filed that note against the front bed. The check we had just added let it through and was right to: the crop matched. Carrots are carrots. What it could not see was which carrots — the part doing the filing only ever looked at the crop, even though the app knew both bed names the whole time.

Now it only holds off when you actually have more than one planting of that crop. If your carrots are all in one place, "the ones in the back" is just how people talk, and refusing to record it there would make the app tiresome for everybody to protect a case that cannot happen to them. Name the bed — "the back bed carrots are up" — and it records straight away either way.

I said my tomatoes had gone orange and it did not offer to log the harvest. Why not?

It should now, and until 17 August 2026 it often did not. When you describe what you can SEE — "they have gone properly orange" — the chat writes that down as an observation rather than as the word ripe, which is the more precise thing to record and the right thing for it to do. The button that offers to log a harvest was only watching for the word. It now reads the observation too, through your crop's own stages: orange fruit on a tomato lands between ripening and harvest window, and both of those mean pick it, so the offer appears.

There was a second half to this, fixed the same day: the chat sometimes filed that observation under the wrong heading, and the app threw the whole thing away rather than reading it — so the same sentence worked one time and not the next. The app now repairs that on the way in instead of asking the chat to be more careful, which is why it is reliable rather than usually right.

What has not changed is the part that matters: the app still checks its own calendar for YOUR garden before offering. If the plant is nowhere near its harvest window, describing it as ripe gets you no button — the chat does not get to decide that on its own, because only the app knows when your plant was put in.

After it added my plants it asked what kinds they were, then when I planted them. Can I skip that?

Yes — every one of those questions has a "not now, make me a task" button, and that is the point of them.

When the chat builds plants for you it asks, in the same breath, what kinds they are — one question for all of them, not one each; if its first reply forgot to ask, the question follows a moment later as a second line. Under its words the app then asks the next things in its own turn: roughly when they went in, and then whether you have a photo, which lets Rafe fill in the rest. Skip any step and it becomes a task on your list for later, so nothing you did not answer is lost. The reason it asks at all: a plant with no kind and no date gets no jobs and no ready window, and typing all that in later is the annoying part.

I described a bed and the plants in it in one sentence, and I got the bed with nothing in it.

Fixed 2 September 2026. When you describe a new bed and what went into it together — three Cherokee Purple tomatoes in the raised bed by the shed — the chat plans the bed and then the plants, and names the bed as the plants' place by the name you gave it. Until today only a bed already on record, or an internal tag the chat sometimes uses instead of the name, counted as a place a plant could go, so the bed was added and the plants were refused: an empty bed you did not make, which is worse than nothing. Now a place made in the same reply counts by its name, in any case, and the plants land in it.

I answered its question about my tomatoes and got "It came back with nothing. Try asking again." What happened?

That was a fault on our side, fixed 2 September 2026 — and the answer you get now tells you what is actually missing.

The chat plans what to record and writes its reply in one go. When it could not place a plant — you had said what you planted but not which bed — it sometimes sent the plan with no words beside it, and the app showed that blank as an error. Measured on the plainest exchange there is: "I planted tomatoes", it asks what kind, you answer, and two times in three you got the error. Now a reply is never blank after the chat has tried to record something. If it could not place a plant it asks the one thing that unblocks it — "Which bed did the tomatoes go in?", or, when you named a bed it does not have yet, "I don't have a bed called that yet — should I add it, with the tomatoes in it?" — rather than the kind or the count, which it can ask later; if it recorded things but said nothing, you get a receipt naming them; and if nothing was recorded at all it says so in one line instead of "got it". A plant needs a bed before it can be added, so that is the question you will see first.

And from 3 September 2026 the same blank cannot happen after a plain question either. Asking "what did I grow last year?", two runs in ten came back with nothing at all. Rafe had decided he needed to look something up — your finished seasons — and said so in the only way he can, by asking us to fetch it. We were ignoring that request at the last moment and had nothing left to say. Now the thing he asked for is fetched and he gets one more go at answering, which is where the answer was all along. If that still comes back empty you get the same honest message and the question is not counted against you.

One more thing fixed the same day: when you answer a question the chat asked, it now reads your answer together with what you said before. Until then it compared your reply with your last message only, so "They're Cherokee Purple, in the back bed" was refused as words it could not trace — the very question it had asked. Your earlier messages in the same chat count as yours; nothing is recorded that you did not say.

I asked how my tomatoes did last year and it said it had no record of my garden. It has eight plants in it.

That was a fault on our side, fixed 2 September 2026 — and it was worse than a gap.

When you ask about a plant, the chat looks it up by name, and until today the look-up kept only the plants growing right now. With one tomato in the bed this year, last year's two finished tomatoes fell out of the answer, the note about their blight became unreachable, and the chat — handed an empty list and no reason for it — concluded you had no garden at all. Now a finished plant stays on record beside this year's one, marked finished with its date, so "did the blight get my tomatoes again?" can reach last year's tomato and what you wrote about it — and when the chat reads a plant's journal by its crop, the finished plants of that crop bring their entries along, each under its own name and the date it finished.

The same day: a garden whose every plant has finished is no longer told nothing recorded yet. It is told there is nothing growing right now, and what it grew, by year — which was always there for gardens with a live plant and never for the ones that had wound down.

From 3 September 2026 that history is one line rather than a table. Which years you gardened, how many plants in each, roughly what was in them. The rest — what was harvested, what died, what you pulled out — is looked up when you ask, the same way it always was for the beds and the dates. Rafe holding less at the start of a conversation is deliberate: on a garden with nothing growing, the list of finished plants is the only thing in front of him, and he will talk about last year instead of adding the plant you just named.

Can I tell Rafe how I garden — that I never spray, that my back bed floods, that my spot runs late?

Yes, from 2 September 2026 — one paragraph, in your words, under Your garden on the account page.

The garden itself is most of Rafe's memory: what is planted where, what you logged, what he read from your photos. Four things do not fit a plant or a bed — a preference (I never spray), a constraint (I travel every August), a correction (you keep telling me to plant out too early for my spot), and standing context (the back bed floods). Write them in How you garden and Rafe reads them on every message, and follows them over his own advice. The spray answer you gave when you set up the app is read the same way.

Three promises, and they are built in rather than promised: Rafe only ever knows what you wrote there — he never guesses about you or writes to it on his own; you can see and change all of it in one place; and when you clear something it is gone from his very next message, because he reads that paragraph fresh each time rather than keeping a copy.

I told the chat I planted something "on Saturday". Does the plant get the right date?

Yes, from 2 September 2026 — and before that it usually got none.

Until then the chat was asked to turn your words into a date itself, and it was bad at the arithmetic: told Saturday, it would pick a Monday, and a guard we had put in for exactly that refused the wrong date — so the plant ended up with no date at all, while the chat read the wrong one back to you. Now the app does the sum, not the chat: it knows today's date where you live, so Saturday, yesterday, two weeks ago and April 3 all land as the day you meant. Words nobody could pin to a day leave the date empty rather than wrong — you can set it on the plant's page. A date still ahead of today is different again, from 16 September 2026: that lands too, as a plan rather than a planting — see the next question.

Can Rafe plant something for a day that hasn't come yet?

Yes, from 16 September 2026. Say "put lettuce in the porch pot on October 6" and nothing goes in the ground yet — it's recorded as a plan for that day. A round + on Calendar starts that sentence for you — tap it and the chat opens with "Plan " already typed, cursor waiting. The reply says plan lettuce for Oct 6 in the pot, and it shows up on your season ribbon outlined until the day comes and you confirm it. Before this, a day still ahead was refused outright: a planting dated into the future would have started counting down jobs for a seed still in its packet. Ask later what is growing and Rafe reads it back as planned, with a countdown — never as a plant already in.

I told it the kind of one of my lettuces and it offered to add another lettuce. Why?

A bug, fixed 2 September 2026 — and the reason the fill card had never once been used.

When you said the kind of a plant you already grow — my lettuce is a Little Gem — the app should have offered to write that on the lettuce. It only could when you grew exactly one lettuce; with two it had no way to ask which, so it fell through to offering a third. Now it asks: Which lettuce is this about? — one tap, then the fill. A kind, a placement or a mulch said about a crop you already grow is taken as describing what you have; a sowing date or a count still means a new plant went in, and that still offers to add one.

There is a small ring beside the mic. What does it show?

Credits, only when you tap it — from 23 September 2026.

On a paid plan, you get 75 credits every Monday. Anything you do not spend carries over — up to 300 in all — so a quiet week banks credit for a busy one, and it never just disappears. The ring beside the mic fills with what you have used; tap it for This chat (what this conversation has spent) and This week (this week's 75, plus whatever is saved on top, and when the next 75 lands).

About what things take: a question usually takes well under 1 credit — most take about 0.2 — a photo check about 2, a step of a garden walk about 2. The work the app does behind the scenes is free and never touches your credits.

Tap "See where they went" to see what used them — Talk, photo checks, garden walks, plants added from tags — each with a count and about how many credits it typically takes.

On a paid plan, if you run out, Talk is paused until Monday. On the free plan, your 5 one-time Talk credits don't reset. Either way, a message you already sent is answered in full; Talk pauses after it. Nothing else in the app stops. Your plants, your dates, your journal and your reminders all keep working.

In the evening the chat said I planted something yesterday when I did it today. Why?

A real bug, fixed 17 August 2026, and it was worse than a wrong date on screen — it reasoned from it. Our server works in UTC, and after about six in the evening in Colorado it is already tomorrow there. So the chat was told the wrong day and then explained your own garden back to you from it: things you did this afternoon became "yesterday", and one afternoon could be split across two days, with a watering at 4:27 reading as today and one at 7:22 as tomorrow. Every interval it quoted was out by one as well — days since planting, days to harvest, days until frost.

Your phone now tells us which timezone you are in, so every date in the conversation is the date on your wall. "Two days ago" also means two dates ago rather than forty-eight hours, which is what people actually mean. If you ever catch the chat and the This week screen disagreeing about what day it is, tell the chat — that reaches us now.

The chat and the This week screen said different things about whether my tomato would ripen. Which is right?

They cannot disagree any more, and until 17 August 2026 they could. The app works out, for every plant you have, whether it finishes before your frost — using your dates and the day you sowed it. The This week screen was reading that answer. The chat was not: it was handed the raw growing times and did the sum itself, which is how it once said "plenty of time" about a tomato the screen had already called too late. It is now given the same verdict the screen shows you, word for word, and told it is settled. If you ever see the two disagree again, that is a bug worth telling us about — say so in the chat and it reaches us.

It told me my tomato will not finish, then said I could pick some before the freeze. Which is it?

Neither half was wrong on its own and the pair could not both be true — fixed 17 August 2026. When the answer is no, the app now supplies the honest consolation itself: they will keep growing and the app will keep tracking them, but the season runs out before they finish, unless you can give them some protection. Those are the same words the This week screen shows you, and there is deliberately nothing after them about a partial crop — if the first fruit is due after your frost, there is no early handful to promise.

The reason it happened is worth knowing, because it is a general risk with any assistant: a flat no with nothing after it is an empty space, and something will fill it. Ours filled it with an encouraging sentence that its own arithmetic, one line earlier, ruled out. The fix was not to forbid softening — it was to put something true in the space.

I added seeds for tomorrow, then told the chat I planted them. Does it know that is the same plants?

Yes, from 20 August 2026. When you add a plant with a date in the future, the app treats it as a plan rather than a record — it waits, then asks on the day whether it happened. Telling the chat "just planted those seeds" now answers that question directly, instead of offering to add a second set of the same plants.

Why it matters more than a tidy list. Every date the app gives that plant — when to expect it up, when it is ready, when to feed it — is counted from the day it actually went in. Until something confirms that day, all of those are guesses. And if it had added a second set, you would have had two records for one real thing, with the reminders arriving twice.

If you said how many, it takes that too. "Planted four of those seeds" puts four on the plant if no number was there before. It will not overwrite a number you typed yourself.

It is the same button as the "Did it happen?" card on Today, so answering in one place clears it from the other.

In Talk, I asked about a plant that isn't in my garden yet. Does Rafe just answer?

He answers, and from 3 September 2026 he offers to add the plant.

Say "I gave my rhubarb a good soaking this morning, was that too much in this heat?" when there is no rhubarb in your garden. You get the answer — that is what you asked — and underneath it, an offer to add the rhubarb. Naming a plant like that is the clearest sign there is that you have one, and it used to be spent on the answer and thrown away.

Rafe already knew. He would say "rhubarb isn't in your record yet, so I don't have notes on it" and then offer nothing, because there was no way for him to hand that over. Now there is.

Asking about a plant is not the same as having one. "How do I grow rhubarb? I have never tried it" gets an answer and no offer — we checked that specifically, because an offer to add something you never claimed to own is more annoying than a missing one is useful.

I asked about a plant I already have and it offered to add it to my garden. Why?

It should not, and from 20 August 2026 it does not. It was only checking whether you had that plant's page open on screen — so asking from the chat, where nothing is on screen, looked to it like you were talking about a plant it had never heard of. It now checks the garden itself.

Two things it goes on, because your name for a plant is often not the catalog's. One is the crop, so asking about your lavender finds your English Lavender. The other is what you called it: we hold Bush Bean as a crop and no wax variety at all, so a plant you labeled Wax Bean is only findable by your own words. Both are checked.

Saying you planted something still adds it, even when you already grow that crop — a second row of beans is a second row of beans. The offer only goes away when you merely mention a plant you have, which is what asking a question about it looks like.

I asked how to start a plant and it said we had not written that up. Had we?

For some plants, yes — and that is fixed as of 20 August 2026. Our plant records hold how each of the 152 crops is started, sourced from US extension services, and none of it was being handed to the chat. Somebody asked whether sweet potatoes with shoots on them could be planted. Ours says sweet potatoes are grown from cuttings, and shoots on a stored root are exactly what those cuttings are. It had the answer and could not see it.

A second thing went with it, and it was bigger. When you named a plant, it only recognized the exact name in our list — so "tomato" worked and "tomatoes" did not. 136 of our 152 plants could not be found by their own plural, which meant a question like can I still grow tomatoes this year reached the part that answers it without knowing which plant you meant. Both plural and singular now work everywhere.

One limit worth knowing: what it tells you is what our record holds, not everything that is possible. A method we have not written down means we have not written it down — never that it cannot be done, and it will say so rather than implying otherwise.

Can it tell me which varieties would still ripen before my frost?

Yes, from 20 August 2026. Ask something like are there faster tomato varieties that would ripen in my season and Rafe counts the days between today and your first frost, then lists the ones that fit, fastest first, with how many of the whole list that is.

The important part is a catch we had to design around. A variety's day count is measured from one of two starting points — from sowing a seed, or from setting a young plant out — and which one it is varies by variety, not just by plant. A third of our plants have both kinds among their own varieties. Ranking those together would put a 50-day figure that starts at sowing next to a 50-day figure that starts weeks later, and confidently recommend the wrong one.

So when a plant is measured both ways you get two lists, never one, each saying what it counts from. Tomatoes happen to be measured one way throughout, so that question gets a single clean list.

And a day count is a catalog figure for a good season, not a promise. A cold August slows everything down, and he will say so rather than letting the number stand on its own.

I told the chat my seedlings have their proper leaves. Did that do anything?

From 20 August 2026, yes. Say what you see and the app writes it on the plant and puts back any jobs that were waiting on it — the same as tapping yes on the card on Today. Two doors to the same place.

It was not working before that, and it is worth being straight about it. The feature was built and shipped, and it needed the conversation to be attached to one particular plant — which nothing in the app ever did. So it quietly did nothing every time. It now works out which plant you mean from what you actually said.

And when it cannot tell, it asks. If you have two sowings of beans and you say the beans are flowering, it will not pick one. It asks which, by name, and writes it down once you say. Guessing would put a job back on a plant that never showed anything, and tell you it had done the right thing.

If you mention a plant you do not have, nothing is written. That is deliberate too — it would rather record nothing than record something true about the wrong plant.

Is it too late to sow something? Can Rafe tell me the last date?

Yes, from 20 August 2026. Ask is it too late to sow radishes and he counts back from your first frost by how long that plant needs, and tells you the date and how many days are left.

Why it works this way rather than from a planting calendar. A single fall sowing date only means something where the fall is long. The published tables we hold are one southern state's, and printing them at somebody in Minnesota would be worse than saying nothing. Counting back from your own frost date works wherever you are.

Seven plants get a straight “we cannot tell you that”, and it is worth knowing which. For artichoke, broccoli, cabbage, cauliflower, sage, sweet potato and tomato, every source we hold measures the days from setting a young plant out, not from sowing a seed. Working a sowing date back from that number would need us to guess how long the seedling stage takes, and being wrong there costs you a season. So he gives you the set-out date instead and says plainly that he cannot give you a sowing one.

And for things you dig or cut — radishes, greens, carrots — he says that sowing past the date still gives you something smaller, rather than nothing. That date is not a cliff for every plant.

What does Rafe say when it is genuinely close?

Both halves, and on purpose. It can work, and an early frost ends it. A plant that needs nearly all the time left before your frost is a real gamble rather than a yes or a no, and picking the cheerful half of that is how you end up covering plants in October that were never going to make it — or pulling ones that would have. The honest answer to a close call is that it is close.

There is a conversation in my list about one of my beds and I never started it. What is that?

The app looking at a photo you uploaded. When you add a picture of a plant, it has a quick look and writes down what it can see — the stage it is at, whether anything has changed since the last picture of that spot. Those notes go into a conversation named after the bed, so everything about that bed stays in one place instead of making a new chat for every photo. It will not message you, buzz your phone, or take you anywhere. It is a notebook, not a conversation waiting for a reply. You can open it, read it, and ignore it entirely.

Can I turn that off, or delete one?

You can delete the conversation like any other, and it will start a fresh one for that bed next time you add a picture there. There is no setting to stop it yet — and we would rather say that plainly than pretend there is. It only ever looks at photos you have already chosen to add, and it only ever writes into that one place.

Does it use up my questions?

No. Looking at a photo you uploaded is not a question you asked, so it is not counted against anything. It is also a much smaller job than a conversation — it is only being asked what it can see, not to talk to you about it.

What happens if I type a question before signing in?

You can't, because the app needs an account before it opens. Using GardenTrack takes a free account with your email, and Rafe is part of the app. If you are ever signed out partway through a question, pressing send takes you to sign in. Your question is kept while you go, and it is waiting in the box when you come back. You press send yourself.

Do I need an account to ask a question?

Yes. The whole app needs one, and it can be free. You make a free account with your email, and then everything opens: your garden, the journal, photos, hardening off and Rafe. Once you are signed in, most of it works with no signal too. Rafe is the exception to that, because the answer is worked out on our side, against the copy of your garden in your account. The only things that work with no account at all are Add to my garden on our plant pages and the free Season Recap, and both of those are on the website.

The chat gave me a different ready date from the one on the plant's own page. Why?

It should not have, and from 13 August 2026 it does not. The plant page used the figure for the variety you picked — Cherokee Purple, 80 days — and the chat used the range for the crop it belongs to, tomato, 52 to 90 days. Same plant, same afternoon, two answers: ripening 1 November on one screen and typically ready 4 October to 11 November on the other. The chat was throwing away the variety you took the trouble to choose, which stings more than it sounds like: this is the second time in a week we have lost that same choice, the first being a bug that never carried it off your phone at all. Both read the variety now, and both fall back to the crop range when we hold no figure for the variety. It is worth more than tidiness. Across the catalog the middling gap between varieties of one crop is about 30 days, and at the extreme it is Purple Sprouting broccoli at 220 days against Melody at 24 — the same crop record. One thing it still refuses to do: if a variety's figure is counted from sowing and its crop's is counted from transplanting, it keeps the crop range rather than convert. Those two differ by the entire time a plant spends indoors, six to eight weeks on a tomato, and a converted guess presented as a variety's own number is worse than the honest range.

I told it I planted something and got a button with no words. Was it ignoring me?

It was not, and it should never have looked that way — fixed 24 August 2026. When you tell the app something rather than asking it something, two things happen: the assistant writes you a reply, and it quietly files what you said so the right buttons appear. Occasionally it did the filing and skipped the reply. You would see a card offering to add your plant and not a single word above it, which reads exactly like being ignored by something that is supposed to be listening. The app now checks before your answer reaches you: if there are no words, it writes a short line itself from what it just filed — Written down — watered the tomatoes, or Got all 3, check them over below. It is a floor, not a substitute. A one-line receipt is worse than a real reply, and the assistant is still expected to write one; this only makes sure a silence never reaches you. The odd part is that half the app already did this. The check existed on one of the two ways an answer can arrive and not the other, so whether you saw words depended on which path your question took — something no gardener could have known or reported.

It told me my beans would not beat the frost, and the dates it gave me say they will.

That was real, and it was our fault rather than the assistant being careless — fixed 21 August 2026. The app works out your ready dates in code, from your own planting date and a sourced day-count, precisely so nobody has to estimate. It then handed those dates over with the instruction never do this arithmetic yourself — and no answer to the one question that matters. We only ever spoke up when the frost won. When the plant won, we said nothing, so the comparison was left sitting there undone. Given two dates, told not to compute, and given no result, it computed, and got it backwards. Now the line says how it comes out, every time. Comfortably clear, and it says how many days clear. Tight, and it says so rather than promising. Ending on your frost date, and it says that in those words. Straddling it, and you get both ends of the window with the judgment handed to you, because a range that starts before your frost and finishes after it genuinely depends on the autumn. And it does not round a hair-thin margin up into a yes. A day-count is a catalog figure for a good season; a cool month moves it, so anything inside a week of your frost is described as tight, not as safe.

It offered to put something on my list without me asking. When does that happen?

When the answer told you to do something — new on 21 August 2026. Before this, the app only offered a reminder if you asked for one, or if it had spotted a pest or a disease worth a second look. So a beta tester who was told to shade his potted beans from 2 to 5 PM through a heat spell had nowhere to put that, and said so: this would have been a great place for it to remember to remind me later. The test is whether the answer told you to DO something, not whether it put a date on it. Feed it, shade it, net it, repot it — those are jobs, with or without a time, and each can become a dated task in one tap.

From 3 September 2026 the task itself is two or three words. Shade tomato. Water roots. Keep picking. Not a sentence explaining why — you would be reading that same sentence every day for two weeks. Nothing is thrown away: the reason, the timing and anything else Rafe said about it are inside the task, one tap in. That holds for reasons you give too — tell Rafe remind me to water Thursday, I'm away for the weekend and Thursday's task says water, with your own reason waiting inside it. If a time was given, the task carries it; if not, it does not invent one. What it deliberately will not do is offer on every answer. Beans like full sun is information, not a job, and gets no button. We checked that against the real assistant rather than assuming: an offer under every reply is worse than no offer at all, because it is the kind of thing you stop reading.

From 2 September 2026 that includes the routine care he tells you to keep doing — pick every two or three days, water at the base, not the leaves — one offer per thing, never one per sentence. He still only offers when his own answer told you to do something; an answer that explains or reassures offers nothing.

It says it has no timing for my orange tree. Do you not have data on fruit trees?

We do, and saying otherwise was our mistake — fixed 20 August 2026. Until then, asking when things would be ready could answer no maturity window in our data for a fruit tree, a citrus or a lavender. That reads as a hole in our records, and it was the opposite: somebody had sourced that plant, decided that a days-from-sowing figure is meaningless for it, and written down why. A home orange is a named variety budded onto a rootstock — nobody grows one from a pip and gets the same orange — so counting days from a seed would answer a question no gardener is asking. We checked how far this went before changing anything. Of our 152 plants, 41 carry no day count. Thirty-nine of those say plainly that the field does not apply, and thirty-nine of them hold the answer in the form perennials actually use: a picking season. The two left over are cover crops — ryegrass and vetch — which nobody harvests. So nothing was missing. The question was being asked of the wrong field. What you get now is the season. For an apricot, when apricots are ready around you; for a navel orange, a winter window that lands in the following year and says so. And it is careful to say what it is not. A picking season repeats every year and is a fact about your area, not about the age of your tree — so it is never handed to you as a first-harvest date. The age is answered separately, and from 21 August 2026 it is answered properly. Until that day the app said it did not know how many years a young tree needs, which was our mistake twice over: we had held that figure for years, and it was sitting in the same records the season came from. An apricot bears 2 to 5 years after planting, a pear 3 to 10, and a lemon grown from a pip 7 to 10 against the 2 to 3 of a grafted one — so the answer now says the figure and where your plant stands against it.

I asked when something would be ready and got "a couple of weeks". Can it do better?

It can now, for most crops. We hold a sourced days-to-maturity window for 111 of our 153 crops — a range like 52 to 90 days, with the thing it is counted from written down, because that matters more than the number. Fifty-two days from sowing and fifty-two days from transplanting are different dates, weeks apart, on exactly the crops people care about. The app now does the arithmetic itself and hands the answer over as real dates: your own planting date plus that window. So you get "typically ready 5 September to 13 October" rather than an estimate. We deliberately do not let the AI do the sums. Adding fifty-two days to a July date is the sort of thing a language model gets right most of the time, and most of the time is not good enough for a date you might plan around. Two honest gaps. Forty-one crops have no window in our data, and a plant with no planting date recorded cannot be worked out at all — in both cases it will tell you that rather than guess.

And since 17 August 2026 it picks the right one of two figures, which on some crops is a month apart. A pepper takes 100 to 120 days from sowing a seed, or 70 to 85 from setting out a plant you bought — the gap is the six or seven weeks somebody else spent raising it. We hold both numbers for 56 crops, and the answer used to take whichever one the crop record listed first, for everybody. Now it uses what YOU did: sow it and you get the from-seed figure, buy it and you get the from-transplant one. Nothing in the plant data could have fixed this — which figure is right is a fact about your planting, not about the crop, so only the part of the app that knows how you started it can choose. The answer says which one it used, so you can see the reasoning and not just a date.

It offered to write down a pest it had just reminded me about. Why?

That was a bug and it is fixed as of 13 August 2026. Somebody asked when things would be ready to pick, and the answer helpfully mentioned the codling moth they had logged the week before — which is exactly what the app is for. Then a row of buttons appeared underneath offering to record a codling moth, as though it had just been spotted. Remembering something is not seeing it. Your journal is handed to the AI so it can answer using your own history, so naming something out of it is really quoting you back to yourself — and offering to save that would file the same sighting twice, on a date it never happened. The buttons now only appear for something it can actually tell from the message you just sent or the photo attached to it. If it genuinely spots a new pest in a new photo, you still get the button.

Why do answers sometimes have stars and dashes in them?

They should not, and from 13 August 2026 they do not. The chat screen prints an answer exactly as it is written, character for character — there is no formatting layer, on purpose, because adding one would mean shipping a sizeable chunk of extra code to every phone just to turn stars into bold. So when the AI reached for formatting, you got the punctuation instead of the effect: two stars, a word, two more stars. The fix was to stop it reaching for them, which also fixed the thing that actually bothered people. Bold headings over short sections turn a two-sentence answer into a page, and you end up reading all of it to find the part that mattered. Answers now come back the way somebody would say them out loud: a few short paragraphs, and then it stops.

The mic stops listening when I pause. Can I make it wait longer?

No, and that is the browser rather than a setting we have hidden from you. The part of the browser that turns speech into words has eight things you can set — the language, whether to show words as they come in, a few others — and not one of them is how long it waits during a pause. How patient it is, is up to your phone. On Android specifically, the setting that would keep it listening is documented as having no effect at all. So we stopped pretending it does not happen. When it stops, the box stays open, your words stay in it, and the button says Keep talking — tap it and carry on, and what you say next is added to the end rather than replacing it. We could have made it restart itself, and chose not to: Android plays a beep every time the microphone opens, and that beep cannot be turned off by a website. Somebody pausing to think would be beeped at for thinking. A button you tap when you are ready is better than a noise every time you stop to look at a plant.

Does the chat know which day of hardening off I am on?

Yes, and until 12 August it did not — it asked you. Somebody told it they had taken their plants out and got back that it could not look up the hardening-off schedule, and then which day are you on? — about a ten-day run the app itself was driving for those exact trays. The app was keeping count and the chat could not see the count. Now it is handed the same thing the Harden screen shows: which day of how many, what today's step is, how long they have been outside today, and how cold is too cold for that particular crop. Ask which day am I on and you get a number. It comes off your device rather than being worked out again on our side, which is the reason the chat and the Harden screen cannot disagree — there is one count, and the chat is reading it rather than guessing at it.

Can the app stop me recording something that could not have happened?

It can now, for five things that are physically impossible rather than merely unusual. You cannot harden off a plant you have said is in a bed — it is rooted, you cannot carry it. You cannot log something onto a plant you have recorded as finished (a note is still allowed, because pulled it out and the roots were knotted is worth keeping). You cannot date something in the future, though we allow a day's slack so a gardener in New Zealand tapping today is never refused. Sprouting cannot be dated before the seed went in — by DATE, not by the order you type things, so catching up on Sunday about Thursday is fine. And an amount needs a unit: a bare 4 against a watering is four gallons or four minutes forever, and the record cannot say which. Three more rules were proposed and thrown out, which matters more than the five: a plant can go backwards, because one that is cut back or hailed genuinely does; a harvest needs no flowers first, because most crops are leaf and root; and a move need not say where it went yet, because forcing that answer makes people invent a bed. A rule that fires on real gardening gets switched off, and then it protects nothing.

If I tell it a plant died, does the plant actually go away?

Yes, from 11 August 2026 — and until then it did not, which was the bug. The It died button on a plant's page closed the plant out properly. Typing my basil died into the chat wrote the sentence into your journal and left the plant alive on your list, still being handed watering jobs for something you had just said was gone. Two ways to say one thing and only one of them was heard. Now both close it the same way, dated to the day you said it happened rather than to today, so catching up on Sunday about a Thursday still reads Thursday. It only acts on a card you pressed Save on, naming the plant — never off a guess about a sentence. If it got the wrong plant, Reopen on that plant's page brings it straight back.

It asked me to photograph something and I said remind me later. What happens when I do it?

You get taken back to the conversation that asked. Finish the job on your list and there is one extra button — back to the chat — and tapping it opens that thread with your photo already attached, under a line saying which day it was and what you were asked. So a question you started on Tuesday and answered on Friday carries on where it stopped, instead of you having to explain it again from the beginning. Two things it deliberately does not do. It does not throw you into the chat the moment you close the job, because you were partway down a list and being moved somewhere you did not ask to go loses your place. And it does not send the photo for you. It sits in the box waiting for you to tap send, because a message that appears in your voice should be one you actually sent. And if you finished the job without taking a picture, the way back is still there — you tap it and the thread opens saying pick up where you left off, because the answer to what it asked may be something you can just type. Closing a job with no photo means no photo, not no conversation.

And from 4 September 2026, finishing a job you asked for writes it in your journal — Done: photograph the underside of a leaf, on the plant, with the photo if you took one. Before that the job was marked done and the journal never heard about it, so Rafe could not see that it happened.

Do I have to say it the way you say it?

No, and that is deliberate. You can name a job by the thing you touched rather than by a gardening verb — "pulling weeds", "threw a sheet over them", "stuck a cane in next to it", "put compost round them", "snipped the tops off" all land on the right entry. You can say it mid-job as easily as after: "watering the beans" works the same as "watered the beans", and "my lettuce is going to seed" the same as "gone to seed". And if you learned to garden somewhere else, we understand that too — "earthing up the spuds" is hilling potatoes as far as the app is concerned, even though we would never write it that way ourselves.

I said I picked my first tomato and it recorded the plant fruiting. Why?

That was a real bug and it is fixed. "Picked the first tomato" was being written down as the plant SETTING its first fruit rather than as you harvesting one — so a moment worth remembering was filed as the wrong thing entirely. The cause was that we had a short list of fruit words — fruit, tomato, pod, bean — that meant "the plant is fruiting", and that list beat the word "picked" sitting right next to it. Which is why "picked the first strawberry" worked fine: strawberry was not on the list. Now the verb decides. If you picked it, pulled it or harvested it, that is a harvest whatever the fruit is called — including fruits we have never listed. If there is no picking word, "first tomato" still means the plant is setting fruit, which is also worth recording.

My sunflower flowered and the app called it a problem. Why?

It should not, and it no longer does. Bolting — a plant shooting up and flowering — ends the crop on a lettuce, a spinach or a cabbage, so the app treats it as bad news. On a sunflower, a cosmos, lavender or chamomile, flowering is the crop arriving, and on a cover crop it is the signal to cut it down. The app now checks what the plant is actually grown for before deciding whether flowering is good news, using what it already knows about that crop. 28 of our 152 plants are grown for their flowers. The reverse holds too: if a lettuce flowers, that is bolting under another name, and the app says so instead of congratulating you.

What can I just say to it, and what does it not follow yet?

You do not need the gardening word for it. "Put some radish seeds in", "put straw round the apple tree", "cut some branches off the plum", "the beans have come through" all land the same as the proper terms — we went and checked how people who have never gardened describe these jobs, and taught it those. Most of the ordinary jobs: planting and sowing, watering, feeding, pruning and pinching and deadheading, thinning, weeding, staking and caging, covering and uncovering, hilling up, mulching, dividing a clump, taking cuttings, harvesting, moving a pot, hardening off, and treating something — spraying, traps, picking pests off by hand. It also understands what the PLANT did, not just what you did: "beans are up", "lettuce has bolted", "first flower on the zucchini". We measured this rather than guessed: 56 real sentences, 80% understood today — it was 57% a week ago and 43% before that, and once you have answered the one question below it is 84%. Anything it does not follow is still saved in your own words as a note — nothing is lost, it just does not count towards a total yet. There are two it asks about instead of guessing: "planted out" and "set out" mean the transplant to some gardeners and carrying the tray outside for the day to others. It asks you once — in the ground, or out for the day — and then it knows what you mean forever, because that does not change between Tuesday and April. Until you answer, those save as your words. It also stopped caring where you put the little words. "Laid down woodchip" used to become a plain note while "laid woodchip round the beds" was logged as mulching — same job, same material, and the only difference was which side of it the word "down" sat on. Now both count, along with "put down bark", "tipped a barrow of compost out" and "chucked down some grass clippings". "Got our first zucchini" counts as a harvest. Nobody says picked, harvested or dug about their first one, and that sentence used to record nothing at all — or worse, record the plant setting fruit rather than you picking it. Not the two things you also get your first of that are not a harvest: a frost, and a flower. And you do not need to know what is eating it. "Something is eating my kale", "bugs on my beans", "holes in the leaves" and "the leaves are getting eaten" now count as a pest sighting on that plant, with your words kept as written. They used to be plain notes, because the app was looking for the NAME of the pest — which is the one thing you do not have when you are asking. What it will not do is invent a sighting from you eating your own dinner: "we ate the first tomato" is not a pest, and neither is digging holes for the tomatoes. And there is an honest floor to this. "Cleared the bed" and "emptied the pot out" stay notes on purpose, because a bed is a place rather than a job — clearing one could be weeding, digging, or pulling out a crop that had finished, and we would rather keep your words than pick one and be wrong.

Can I ask him about the app itself?

Yes. Ask how your plan works, whether it needs signal, where your photos are kept or what changed recently, and he answers from these help pages — the same words you would read here. If we have not written a page about what you asked, he says so instead of guessing. That is deliberate: he can only tell you things we have actually written down and checked, so the worst he can be is out of date, never invented.

What happens to a question it could not answer?

It becomes the note about which help page we write next. If you ask something about the app and there is no page for it, your question is passed to us in your own words — not a tally, the actual sentence you typed, because that is the thing worth reading in three months. You do not have to do anything and it does not tell you it has done it. One person asking once is just a question; the same question turning up three times is a page we owe you.

Will asking about the app confuse it about my garden?

No, and the two are kept apart on purpose. "Where are my photos stored?" is about the app. "Why does it think my tomatoes need water?" only sounds like it is — that one is about your plant, its dates and your own notes, and it is answered that way. If a question genuinely has both in it, you get both answers and it says which is which.

Rafe offered to make a bed. How do I say no?

There is a button for it, and from 27 August 2026 it actually looks like one. When Rafe wants to make something you get the thing he is proposing and three buttons, stacked: Make the bed, then Make it, and stop asking me, then Not the bed. Tap any one. Nothing is created until you do.

The no was hard to see, and that was our fault. It was drawn with an outline so faint you could barely make it out against the page, and it sat underneath a full-width orange button — so it read as a caption rather than a control. On a screen whose entire job is asking your permission, the half that declines was the half that did not look tappable. All three now sit full width, one under the other, and all three look like what they are.

None of them is a plain Yes or No, and that is deliberate. They say the bed or the plants, because the message above them can also contain things Rafe has already written down for you — Already logged, watered the tomatoes. A button reading just No next to that sentence could reasonably be read as undoing the watering. It never would have, but you would have had every reason to think so, and being wrong about what you just did to your own records is worse than an extra word on a button.

The middle button changes a setting as well as making the thing — it tells Rafe he may create things without asking you again. You can put that back at any time under how much Rafe may do, on your profile. Saying no does not make Rafe stop offering and does not delete anything you already have. It simply does not make the thing.

I said I built a bed and planted things in it. Why were the plants not in the bed?

They are now. Say "I built a new raised bed and put a row of lettuce, spinach and kale in it" and you get one bed with three plants drawn inside it, straight away — you do not have to answer the size question first. That used to arrive a turn late: the first card showed three loose plants and a Save all button, and the bed only appeared after you said how big it was. Anyone who took that button ended up with three plants sitting nowhere and a bed that had to be joined to them afterwards. The two are different kinds of thing. That the plants went in the bed is something you already told us; how big the bed is, is a detail we can ask for after. Waiting for the second before showing the first meant offering you a version of your own sentence that was missing half of it.

It told me a crop would not finish. Does that really mean I get nothing?

It depends on which part of the plant you eat, and from 14 August 2026 the answer says which.

For anything you dig up, running short of days means a smaller crop rather than none. Sweet potatoes want about 90 days for full-sized tubers. If you have 61, the tubers keep growing right up until the frost kills the vines — and killing the vines is when you dig. So what is in the ground that night is what you get: small, and real. The same is true of anything you cut: baby kale is a crop.

For anything that has to ripen, it really does mean nothing. Green tomatoes do not ripen after the vine dies, an unfilled pod stays empty, and there is no small version of a broccoli head that never formed. Those still get a plain no.

The app will not make the call for you. Where it is a smaller crop, it says so, says why, and leaves it there — worth doing if you want any at all, not worth it if you were counting on a proper harvest. That is your decision and it depends on things we do not know, like whether you have somewhere better for that bed. This came from a gardener pushing back on a flat no, and he was right to.

The round button in the corner was half off the bottom of my screen.

Fixed on 14 August 2026, and it was covering part of the Profile tab while it was there. That button floats above whatever is at the bottom of the screen, and on the Talk screen it works out where to sit from the height of the typing bar. When that measurement came back wrong it did not just sit in an odd place — it sat below the bottom edge, clipped, on top of the corner of the Profile tab underneath it. Profile still opened from its left side, so it was partly stolen rather than dead, which is the sort of thing you feel as the app being slightly broken without being able to say why.

It now has a floor it cannot go below, set to the same distance it uses on every other screen — the one that clears the tab bar. Whatever the measurement says, a button that is meant to be raised can never end up lower than an ordinary one.

There used to be two microphones on the Talk screen.

There is one, and only one, from 14 August 2026: the microphone in the typing bar. That is the one that dictates your question.

The round floating button — the one for telling us about the app when something looks wrong — is simply not on the Talk screen any more. It sat a thumb away from the composer doing an entirely different job, and briefly wore a pencil to tell them apart, which was still one button too many: the typing bar is already the way in on that screen. It is on every other screen, unchanged, where nothing competes with it.

So if you want to report something about the app while you are in a chat, step out to any other screen and it is one tap. That is a real trade and it is the right way round — the chat is where you talk to your garden.

The chat showed my question twice, once as a heading and once as a message.

Fixed on 14 August 2026. A conversation is named after its first message, which is exactly right in the list of past conversations — it is how you recognize one among twenty. Inside the open conversation it sat directly above the same words in your own message bubble: two identical blocks of your writing before the answer began, which on a phone is most of the screen. The heading now just says Rafe, and the conversation keeps its name where that name does some work.

I asked in one message and answered in the next, and it lost the thread.

Fixed on 14 August 2026. Somebody asked whether sprouting sweet potatoes could go in this year, got a question back, and answered it — and by then the check that works out whether a crop can finish had gone quiet. It was reading only your newest message. On its own, "I have a sweet potato in my kitchen and it is the one growing off shoots" asks nothing, so nothing fired.

It now reads the turn before it as well, which is how conversations actually go: you name the plant in one breath and the details in the next.

Two things it deliberately will not do, both because a stale answer attached to a new question is worse than no answer. It only looks back one turn — six messages ago is usually a different subject. And it only carries on when you are still talking about the same plant: ask about sweet potatoes and then say "what is eating my kale", and you get an answer about kale rather than a planting verdict about sweet potatoes. The first version of this did exactly that, and we caught it before it shipped.

I asked whether I could plant something and it did not seem to notice the question at all.

That was a real hole and it closed on 14 August 2026. The app works out for itself whether a crop can finish before your frost — but only once it recognizes that you are asking. It was checking for a short list of ways of asking: can I, should I, is it too late, when. Somebody who wrote "am I able to plant these this year" — a planting question in plain English — matched none of them, so the whole check stayed silent and the answer came back as ordinary encouragement.

It now recognizes intent as well as questions. "I want to plant this", "thinking about putting in some kale", "planning to sow spinach this weekend" and "is there still time to start beans" all count, because somebody saying what they are about to do has not done it yet — and telling them it will not finish is the entire point.

What it deliberately still ignores is anything already in the ground. "I planted the beans in June, when will they be ready" is a question about a plant that exists, and answering it with no, that cannot finish would be the app arguing with something you watched happen. Measured across 11,166 sentences of real gardener writing, widening this made the check speak up six more times — and it recovered the question it had been missing.

I asked if it was too late to plant something and got a wishy-washy answer. Has that changed?

Yes, from 13 August 2026, and it was a real mistake rather than a vague one. Asked whether sweet potato slips were worth planting in mid-August, the chat replied that there were 61 days until first frost and that this was "tight but workable" for a crop needing 90 to 150 days — then suggested planting them as soon as possible. 61 days against a 90-day minimum is 29 days short of the fastest variety there is. Not a smaller crop. No crop at all. Every number in that answer was right and the conclusion drawn from them was wrong.

The app works it out now instead of the AI. When you ask whether you can still grow something, your frost date and that crop's own days-to-maturity are compared before the question is even sent, and the answer is handed over as a settled yes, no or risky. The reply leads with it and then explains — and when the answer is no, it tells you what you can do instead, like potting the slips up for next spring. Measured on that exact question: the honest answer came back 4 times in 15 before, and 14 times in 15 after.

And the wrong day count is fixed too, as of 14 August 2026 — the cause turned out to be ours. The AI used to write its own countdown to frost and get it wrong, saying 69 days where the real figure was 61. We had assumed it simply could not do the subtraction. In fact we were handing it a sentence beginning "Only 61 days remain..." and then, two lines later, telling it never to state a number of days. It was copying the shape we had shown it. The answer now names the frost date instead — the first frost comes Oct 13 — which is exact, checkable against your Today screen, and nothing has to be worked out. Measured on the same question: a wrong countdown in 15 rolls out of 15 before, and 0 out of 15 after. It can still occasionally slip one into an answer about a crop you dig up, which we are still on.

Can I fill in a detail about a plant I already have, just by saying it?

Yes, from 13 August 2026 — four things: which kind it is, the day it went in, how many, and which bed. Say "I put six Red Russian kale in the back bed on Saturday" about kale already in your garden and one card appears listing exactly what it is about to write, with one button. No going into the plant, no form. Naming the crop is the surest way — the app matches that word against what you are actually growing. But it no longer depends on you saying it. If you name no crop and only one plant in your garden is missing everything you just mentioned, there is nothing left to guess and the card appears anyway. If two plants could be meant, it stays quiet rather than picking one.

Until this week none of that did anything at all. The app could record all four while you were adding a plant, and could record none of them on a plant it already had — so the only way to fill a gap was to go and tap into the plant, which is exactly the tapping the chat exists to save you.

It only ever fills blanks, field by field. Say four things about a plant that already knows three of them and only the empty one is carried; the card cannot become a correction, whichever ones you happen to mention. The date is shown as a real date before you save it — say "Saturday" and the card reads 9 August 2026, because a plant quietly dated three days off is worse than one that asked. Words that name no day at all, like "in the spring", write nothing. More on Which kind is it?.

Two things fixed on 2 September 2026. Saying which kind a plant is used to rename it as well — a plant you had called Wax Bean came back as Bush Bean · Golden Wax. Now the kind is written on the plant and the name you gave it stays exactly as you typed it; only a name the app made up itself (Bush Bean, Bush Bean 2) is filled in. And the card used to vanish after you tapped it with no word that anything happened; it now says ✓ Written on the plant, with a link to it.

And from 2 September 2026 a sentence that starts with a plant’s name is heard as you speaking. “Nan’s broccoli is a Calabrese variety” used to get a summary of what Rafe knows about your garden and nothing written down — your sentence was arriving as the last line of the notes Rafe reads first, so it read as more notes. Put my in front and it always worked. There is now a clear break between what Rafe is told and what you say to him.

What do I get back when I tell it four things at once?

One card, not four forms. Your beds are drawn as beds with the plants sitting inside them — six tomato icons in the back bed, three peppers beside them — and anything that is not a plant gets its own line underneath. One button saves the lot. The point is that you can check it by looking rather than by reading: a pepper drawn in the wrong bed is wrong before you have read a word, where ten lines of correct-looking text just get a tap.

What if it gets one of them wrong?

Tap the row to open it and fix it, or tap the small ✕ to throw that one away. The rest stay. That matters more than it sounds — without it, one invented plant would force you to either accept it or abandon the four correct things beside it, and both are worse than checking a card.

Can I make it stop asking?

Yes. There is a Save without asking switch on the card itself, so you can turn it on the moment you have watched it work and decided you trust it. After that it saves right away — but it still tells you: a one-line "4 things saved" with an undo next to it. You can turn the card back on any time under Profile. A setting you cannot reverse is not a setting.

If I say I watered the cherry trees, does it know which trees?

Yes — it attaches the watering to the actual plants, one entry each, so each tree can answer for itself when you ask what it has had. It matches on the crop, and on the place when you name one. If everything of that kind is in one spot you do not need to say where. If it is not certain, it asks rather than guesses — say "the two cherry trees" when we hold three and you get the three offered to pick from. A number you stated is something you know about your own garden, and a watering attached to the wrong tree is never noticed.

Can I tell it several things at once?

Yes, and that is the point. "I put in six tomatoes and three peppers on Saturday, built a bed out back, and remind me to net the cherries in June" comes back as four cards — two plantings with their counts and their date, the bed, and the reminder — each with its own button. Nothing is written until you press one, and you can press some and ignore the rest.

Will it put the wrong number on the wrong plant?

No, because it will not put a number it is unsure of on anything. When it can tell that the six goes with the tomatoes and the three with the peppers, both cards arrive filled in. When your sentence is genuinely ambiguous, both cards still arrive — with the number box empty for you to fill in. An empty box is a question you answer in a second; a wrong number is an answer you would never think to check.

How is this different from asking a normal AI chatbot?

A general chatbot starts from nothing and answers about the species. This one opens already knowing today's date, your frost dates, every plant you have with the day it went in the ground, and the last dozen things you did in the garden. So 'will the tomatoes ripen before frost?' comes back with your frost date and your planting date, not a paragraph about tomatoes in general. Ask him something he has no note about and he says so rather than making something up.

If I ask what powdery mildew looks like, where does the answer come from?

From text we wrote off a named source — a university extension service or a plant clinic — and he tells you which one, with a link, so you can go and read it yourself. We hold descriptions for 149 of the problems our plants get, which covers a bit over four-fifths of the times a plant in our database names a problem. Ask about one we have not written up and he says so rather than filling the gap from memory. The limit is worth knowing: these say what each problem looks like on its own, not what tells it apart from the one beside it. So if you ask about "mildew" you get both the powdery and the downy kind, and he says plainly that he cannot rank them for you — because telling those two apart from a description is a thing we cannot honestly do yet, and guessing would be worse than saying so. Where we know a problem is tied to one crop, he says that too — the fusarium wilt that kills tomatoes is a tomato-only strain, so if it took your tomatoes it is not coming for your beans next, and that is usually the thing you actually wanted to know. And where the page we learned from covers something narrower than your plant, the answer tells you that as well, rather than quietly applying a pepper study to your beans.

It said it could not see my task list, but Today is showing me jobs. Which one is right?

Today is right, and you should go by it. Your task list is worked out on your phone, and the chat is sent a copy of it with your question. If that copy cannot be built for any reason, the chat now tells you so and sends you to Today — because the alternative is worse. It used to go quiet instead, and a missing list looks exactly like an empty one, so it would answer "nothing is due this week" in a week that was full. A chat that admits it cannot see something is safe; one that confidently tells you there is nothing to do is not. Anything you asked for by name still comes through, because those are yours and do not depend on the list being built.

Does Rafe remember what I said earlier in the conversation?

Yes. You can say 'and which of those is furthest along?' and he knows what 'those' means. He keeps the conversation whole up to a size limit, so a long conversation loses only its oldest end, and only once it runs on for a while — never its most recent turns, which are almost always the part that matters.

Can I have more than one conversation?

As many as you like. Chats, at the top right, holds every past conversation, newest first, named after the first thing you said in it. Tap one to pick it back up, or start a new one. They stay there, so the conversation about the peach in August is still there next August.

Can I send more than one photo at a time?

Up to four in one message, and he reads them as one set rather than four separate questions. That matters more than it sounds: the top of a leaf and its underside are two views of the same problem, and a lot of things — mites against rust, for instance — cannot be told apart from one angle. When you send several they are numbered, so you can say "the second one" and he will know which you mean.

Can I show Rafe a photo?

Yes — the camera button in the message bar takes one, and the button beside it picks one you already have. Attach a photo HERE, in the chat, and he looks at it straight away — handing a picture to a conversation is itself the question, so he does not stop to ask. (A photo taken on a plant page is the other way round: it asks first. Asking about a photo covers that.) He looks at the picture alongside that plant's own history. A question with a picture in it goes to a stronger model than a typed one, because working out what is in a photograph is a harder job than looking something up in your records.

What are the buttons under an answer?

Next steps the answer opened up. Show him a plant he has no record of and you get "Add this apple to my garden". Show him a pest he can name and you get "Log codling moth on your apple tree" — one tap, no form, because everything a form would ask has already been said. Show him damage he cannot name and you get "Check it for pests and disease" instead, with the exact photos to take. Ignoring them costs nothing, and nothing is saved until you tap.

Why does Rafe offer to log a pest sometimes and to look for one other times?

Because those are different situations and only one of them is a question. If he can name what he is looking at, the useful thing is to write it down. If he can only say something is wrong, the useful thing is a closer look — and he tells you exactly which photo would settle it. He will not ask you to go hunting for a pest he has just identified, which is the sort of thing that happens when a chatbot picks its own buttons. Ours does not: he reports what he can see, and the app decides what that makes possible, using things the chat cannot know — whether that plant is in your garden, whether you already logged this one this week, whether a job is already on your list.

It saved a note and I could not tell which plant it went on, or change what it said.

Both are fixed as of 21 August 2026, and the second one is the bigger change.

The note now opens before it is saved. Tap the plant and a small card appears with the sentence the app is about to write, in full, in a box you can edit. Change any of it, then save. Nothing is stored until you do, and the card says so in as many words — because a card showing a finished sentence is exactly the sort of thing you would reasonably read as done.

Why it needed to open at all: the button could only fit about forty characters, so most of the sentence was never on screen. You were agreeing to words you had not read. And once a note is in a plant's story there is nowhere to edit it, so the moment before saving is the only chance there is.

And the tick afterwards names the plant and links to it — "Noted in Bush bean's story. Open it" — where it used to say "Noted in this plant's story", which names nothing and points nowhere. A confirmation that will not say where something went is the same thing as a note filed on the wrong plant: nothing looks broken and you cannot find it later. The same is now true of a pest, a disease and a harvest.

Saving does not end the conversation. It never did on its own — but the app reloads itself when an update lands, and until now that dropped you back to a blank Talk screen. The conversation you have open is now part of the address, so a reload puts you back in it. Starting a new chat still leaves the old one, exactly as before.

How does it pick the variety?

From a list of the ones we actually hold for that crop — 59 for apple, and it pre-selects whichever one it guessed from your photo so most of the time you change nothing. If yours genuinely is not in the list, there is a link to type it in by hand. That name is saved on the plant but it will not drive your ripening dates, because a variety we do not hold has no dates to give — and we log it so we can go and add it properly. Same if the chat names a variety we are missing: that is the more useful signal, because it means our list is behind what people are growing.

Can Rafe add a plant for me?

Tap "Add this apple to my garden" and you get the details filled in — what kind of plant, what to call it, where it is, the variety if he has a guess. Edit anything wrong and save. Nothing goes into your garden until you do, which is the point: he can be wrong about an apple, and you will see that it is before it becomes a record.

Why does it not just add it itself?

It depends on your setting, and since 31 August 2026 the default is that it does. Tell Rafe you planted something and it goes into your garden straight away; the line under the answer says exactly what was made, and Change or undo takes it back. Set Rafe to ask first, and you get the card instead — the details filled in, waiting for you to tap.

Why there is a card at all: a wrong answer you can see is fine and a wrong record you cannot see is not. He once told a gardener the apple tree was added when he had done nothing at all, and there was no way to tell from the screen. A card cannot fail that way — it is either there or it is not — and neither can the line under the answer, because it is built from what actually reached your records, never from what Rafe meant to do.

Can Rafe remind me to do something later?

Yes, the same way he adds a plant — he offers, you save. When an answer depends on something you cannot check right now, he puts up a card with the job, when to do it, and how to do it properly. Change the day with one tap and add it to your list, and it appears on Today when the day comes. The how is the part that matters: a reminder that says "photograph the leaf" gets you another picture of a whole tree, and one that says "hold it flat, underside up, in shade, fill the frame" gets you an answer.

Do reminders sync to my other devices?

Not yet — they live on the phone you set them on. Everything else in your garden backs up to your account; reminders are new and their half of that is still being built. The app says so on the card rather than letting you find out.

Can Rafe log a job or a harvest for me?

No. Adding a plant and setting a reminder are the only two things he can offer. For anything else he will tell you plainly that he cannot and point at where you can. If he ever claims to have done something in the app, that is a bug worth reporting — he is built not to say that.

What will Rafe not tell me?

He will not name a shop product or a dose. He names what works, sprays included, by the active ingredient the way an extension service does — "horticultural oil", "Bt", "copper" — says when to apply it, puts the fixes that need no spray first, and tells you to check that the product label lists your crop, because the label carries the amount and is the only correct source for it. (Until 5 August 2026 he named no chemical at all; that changed because a fix you cannot use is not a fix.) He will not tell you when something was planted or how old a plant is unless you told the app first. And he will not tell you how a plant looks unless you show him a photo, because a watering note says somebody watered it, not that it is doing well.

How many questions do I get?

On GardenTrack Free, 5 Talk credits, once. That's about two dozen messages. Not 5 a month: 5 in total, counted from the day you make your account, and they never refill. A message usually takes a fraction of a credit, and Profile and the ring beside the mic show what is left to two decimals. The free account's photo taste is separate and works the same way: 2 diagnoses, 2 photo adds and 1 garden walk. On a paid plan, Talk comes out of your weekly credits instead.

Profile tells you where you stand, as a plain count of what is left, and so does the ring beside the mic. If the server reports a different number for your account, that number is what you see, because a fact about your account beats a general statement about the plan.

Why once rather than a monthly allowance: a small number each month teaches you to ration and wait, and a gardener with a sick plant in week two does not ask; an assistant you have learned not to consult is not an assistant. Refusals and errors do not count against you.

And when you reach the end, it tells you plainly, with no Try again under it, because trying again cannot help and a button that cannot work is worse than no button. Writing in your journal never needs AI, so running out never stops you recording your garden.

If I tell Rafe my seedlings have their first true leaves, does that get saved?

Yes, as a stage on that entry — alongside what our model expected on the same date. True leaves, cotyledons, setting fruit, heading up, bulbing, pods filling, starting to ripen, curing, gone dormant all land as observations. Flowering, bolting, germinating and fruiting already had their own entries and still do. What it will not do is take a stage out of a question. "Should I pinch the flowers off?" has the same word in it as "it is flowering", and only one of them is you telling us something. Plans, worries, conditionals and "not yet" all record nothing — your words are kept as a note instead, which loses nothing. A wrong stage in your journal would be permanent and invisible, and a note is neither. One exception worth knowing: "the peppers are setting fruit, is that early?" does count, because you told us something and then asked about it.

Does it know what stage my plants are at?

Yes, and he reads what was written down at the time rather than working it out now. Every entry you log records what stage the plant was in and what stage our model expected on that date. Ask how something is doing and he can tell you both — "watered while it was flowering, where the model expected leafy growth" — and you draw your own conclusion. He will not tell you that you are ten days ahead, because that is not a thing the record can support: the stages are not a single ladder every crop climbs in the same order, and we do not save the dates we would need to count from. Stating both and stopping is the honest version. And when there is no stage he tells you why. We hold no growth-stage model for this crop is a different problem from our dates for this crop did not line up with your planting, and you can act on the difference. Anything logged before 11 August 2026 has no stage on it at all, and we do not go back and guess.

It said it did not know my frost dates, but the home screen shows them. Which is right?

Both were, which is the annoying part, and it is fixed. Until you give a ZIP the app uses the typical frost dates for your zone — the home screen shows them and says they are typical. The chat was never told about them, so it honestly said it had no note of them, and then it explained itself by inventing a setup step you never did and pointing you at a settings screen that does not exist. Now the chat is sent the same line the home screen is showing, and where the dates came from. Ask it today and it gives you the date, tells you it is the typical one for your zone rather than yours, and says that adding your ZIP gets you your own.

If the app is already warning me about something, will the chat still say it is fine?

No, and that is deliberate. When you ask about a plant, the chat is told about anything the app is already telling you to look at on that same plant. It will not agree that it is nothing to worry about, even if you ask it that way — and people do, because "that is just old leaves, right?" is how anyone would ask. It will say it can be more than that, name the two or three things it could be, and tell you the one detail that separates them. Two parts of the same app disagreeing about your plant is worse than either one being wrong on its own.

So does it just tell me to worry about everything now?

No. It only holds the line on signs the app has actually flagged on that plant. Everything else is answered normally, and telling you something is perfectly normal is still the most common true answer there is — about half of all of them. When it does reassure you, it should say what the plant is DOING, not just that it is fine. Being told not to worry is not the same as understanding why.

Can I just tell it what I planted instead of filling in a form?

Yes. Say something like "I put in two rows of Sungold on Saturday in the back bed" and it offers to add the plant — with the crop, the variety, the bed and the date already filled in. Nothing is saved until you check the card and tap the button, and every field on it can be changed first.

From 2 September 2026 the kind of plant is checked against our list before it is saved. Rafe picks the crop from the words you use, and he can pick a name we do not keep — lettuce on its own, when we hold head lettuce and leaf lettuce separately. Until today that plant was saved under a name nothing could look up, so it got no watering, frost or harvest guidance and nothing said so. Now a close match is taken (sweet corn, pumpkin), and one we cannot tell apart is saved under the word you used and marked as a plant we do not hold yet, so its page does not promise guidance it has not got.

How does it know which Saturday I meant?

It works it out from today and shows you the actual date on the card before anything is saved, labeled with the words you used. If you said something it cannot pin to a day — "in the spring", "a while back" — it leaves the date empty rather than guessing, because every job the app ever gives that plant is worked out from that one date. A missing date is fine. A wrong one is not.

I said two rows. Why did it not fill in how many plants?

Because two rows is not two plants, and we have no idea how many plants fit in your row. Rather than invent a number that would sit in your records forever, it leaves the box empty and tells you why. Type it if you know it.

Can it add a bed or a job too, or only plants?

All three. "I built a raised bed out back, it is 4 by 8" offers to add the bed with its size already in. "Remind me to prune the apple tree in December" offers to put that on your list, dated. Same rule every time: it fills the card in, you check it, nothing exists until you tap.

Why does it care how big my bed is?

Because width and length are the only facts about a bed the app will ever need, and once the bed exists nothing will ever ask you again. They are what spacing advice runs on. If you give the size in meters or inches it will leave the boxes empty rather than convert — a bed recorded as 4 feet when you said 4 meters is not a small mistake, it is a garden that does not fit.

I said remind me at some point. Why did Rafe ask when?

Because a reminder that goes off on a day nobody chose is worse than one you have to date yourself — the wrong day is the one you will trust. He will not pick a date for you. Say a month, a weekday, or something like in two weeks, and he will work out the day and show you.

Can I just tell it what I did instead of tapping through the app?

Yes, and this is the one people will use most. "I watered the tomatoes this morning, gave them about two gallons" offers to write that down as a real watering — on that plant, dated this morning, two gallons — not as a sentence you have to re-read in three years. Same as always: it fills the card in, you check it, nothing is saved until you tap.

What if it cannot tell what kind of entry I mean?

It saves it as a note in your own words and tells you that is what it did. Nothing is lost. We would rather file something loosely than guess wrong — "I pulled the dead tomato out" is not a harvest, and a wrong entry is worse than a vague one because it does not look wrong later.

I said I gave them a good soak. Why is the amount empty?

Because there is no number in that. The whole point of recording how much is being able to compare one year to the next, and a made-up figure would quietly break that. Type it if you know it, or leave it — the watering is recorded either way.

I described a bed and the plants in it. Do the plants go into that bed?

Yes. Say "I built a raised bed by the fence and put four tomatoes and some basil in it" and the tomatoes and the basil are filed into that bed, not left loose in the garden. Two cases where it will not guess, on purpose. If you name a different spot for a plant — "the basil is on the kitchen sill" — your words win over the bed you were describing. And if one message sets up two new beds, it files nothing and asks, because a plant put in the wrong bed is worse than one not filed at all: an unfiled plant looks unfinished and you go back to it, while a wrongly filed one looks done.

My own messages were washed out and hard to read. Was that me?

No, it was us, and it is fixed. For a few hours on 13 August the bubble your own words sit in was filled with a pale lavender that is meant for text on a dark page, not for a background — so white words on top of it were close to invisible in both the light and dark themes. Nothing you typed was lost or changed; it was only the color behind it. If you are still seeing it, close the app fully and open it again so it picks up the new version.

I opened my past conversations and the list was blank. Where did my chats go?

Nowhere — they were all still there, you just could not see them. For a few hours on 13 August that panel painted itself the same color as its own writing, so the list looked like an empty rectangle. It is fixed. Nothing was deleted and nothing needs restoring: open the panel again and your conversations are where you left them.

Can I ask it to delete a plant, or change a date?

Both.

Tell him the tomatoes actually went in on the 3rd, that it is a Sungold, not a Roma, that the apricot has died, or that the lavender now lives in the big pot, and he lays the change out under his reply, read back from your records. Ask him to take out a plant, a bed or a zone and he shows the removal as a card: nothing is removed until you tap it. If a bed still has something in it, he says so instead of guessing you meant to empty it first.

Two ways a plant leaves your list, and he uses the right one. It is done — or it died — keeps everything: the variety, when it went in, everything you did with it — and just drops it off your list. Remove is for something you added by mistake, and it goes. A plant that really grew is worth keeping the record of, so say which, and he does that one.

It says it could not tell which plant I meant. What now?

Tap the one you meant, and it goes in. From 31 August 2026, when we cannot work out which plant you were talking about, the app asks — with a button for each plant it might have been, each one named.

Say “I watered the beans” with three lots of beans in your garden and there is no honest way to guess. We will not pick one, because writing to the wrong plant is worse than not writing at all and you would have no way to notice. So you get Not written down yet — which one did you mean? and the three names. Tapping one writes it straight away.

⚠ There is no “yes” button, on purpose. The buttons say the plants’ names. A button reading just Yes answers whatever you thought you were asked — and on a screen that writes to your records, that is not good enough.

If you asked for several things at once, it quotes the bit it is talking about — I could not tell which plant you meant by “I watered the beans” — so a correction sitting under a long answer cannot be read as being about the wrong one. When it is the only thing on the line, it does not bother.

Sometimes there are no buttons, and that is deliberate too. If two plants share a name AND sit in the same bed, nothing we could put on a button would tell them apart — so we say nothing was written and leave it with you, rather than offering a choice you cannot actually make.

I described my plants clearly and it asked me the same question four times, and changed nothing. Why?

Fixed 8 September 2026. That afternoon somebody typed my front porch has two plants in it, the orange tree and the lavender, each in its own pot — the apricot is in the ground, planted in April, and got Not written down yet — which one did you mean? four times over, with nothing changed.

What happened: Rafe looked the plants up to get their exact names, then used those names when he tried to change the records. The app’s rule was that Rafe may only touch a plant you named in your own words. Your words were right there; the names he chose were not, so every change was refused. He tried once more, and each refusal was drawn as its own question.

Now: your words or the plant’s exact name both count, as long as you named that plant in your message — and if you have two oranges, the orange tree still makes him ask rather than guess. The same question is drawn once. Anything Rafe has already looked up in a turn is not looked up again. And tapping a plant under a change he could not make now makes that change, instead of writing an empty note.

How does Rafe know how the app works? Can I ask him about the app?

Yes — and from 8 September 2026 he opens the help page first. Ask whether the app works without signal, where your photos are kept, or what is new, and Rafe reads the page written for that question and answers from it, with a link, instead of working it out from how the app behaves. The pages about the things Rafe works inside — asking, photos, writing things down, your lists, your beds, your dates — he keeps with him at all times; the rest he opens when you ask.

If no page covers your question, he says we have not written it up yet. That is on purpose: a guess about a product sounds exactly as confident as a fact about one, and we would rather you got the honest gap than a confident invention.

I have a lot of plants. Does Rafe still know about all of them?

Yes, and from 2 September 2026 he says so bed by bed once you pass forty. Rafe is shown your living plants one by one up to forty. Past that, he used to be shown the twenty that had been in longest and the twenty planted most recently, and only a count for the rest — so a plant in the middle could drop out of view, and Rafe would sometimes say he had no record of a plant you had. On a sixty-plant test garden that happened nineteen times in twenty.

Now, past forty, Rafe sees each bed with how many plants are in it and which crops — Raised bed 2 (14 plants): tomato ×2, basil ×4, … — plus the plant you are asking about on its own line, and he is told exactly how many are on record. Nothing drops out without being counted, and on the same test the false "no record" fell to one in twenty. Two plants always keep a line of their own above the bed-by-bed list: the one that has been in the ground longest and the one you planted most recently — the tree you have had for years and this spring's sowing are the two a gardener asks about first. Ask about a bed or a plant by name and Rafe looks it up in full.

And from 6 September 2026 he also gets the complete list of what you are growing. The bed-by-bed view above cut the wrong answer down but did not end it — in testing he still said “I don’t have a record of broccoli in your garden” about broccoli in the front bed. He was told his list was cut and simply did not act on it. Now the crops themselves come through complete, beside the shortened list, so a crop you grow cannot read as one you do not — he can check rather than having to remember. It is only sent when the list is actually cut; under forty plants nothing here applies to you.

I told Rafe a plant went into a bed, and he says there is more than one bed called that. What now?

From 3 September 2026, he asks you which one, and tells them apart by what is growing in each. (And if Rafe himself made up a bed name you never said, he now tells you which name he could not place and lists your beds, instead of asking you for the plant and the bed all over again — he only offers to make a new bed when the name was yours.) If two of your beds share a name, the card under the answer lists both — one might say Raised bed 2 — lettuce ×3 and the other Raised bed 2 — empty — and tapping one puts the plant in that exact bed. Before this, both cases came back as I could not tell which plant you meant, which was about the wrong thing and left you nowhere to go. If you name a bed you do not have, he now says so and offers two ways on: put the plant in one of the beds you do have, or make a bed with the name you said and put it there.

Does Rafe know which beds I have, even the empty ones?

From 2 September 2026, yes. Rafe is told how many garden zones, beds and seed trays you have and how many of them are empty, and he can look the whole map up by name whenever a question or a new plant needs it. Before this, Rafe was shown your living plants and nothing else about the places they were in: a bed with nothing growing did not exist to him, a garden zone never did, and a garden of six empty beds was described to him as having nothing recorded yet. That is why he could not say which beds were free. Rafe still does not get the full map on every question, on purpose: when the map was put in front of him every time, he started asking about your beds instead of adding the plants you named. So he gets a signpost, and opens the map when he needs to.

I told Rafe about my beds and everything in them, and he asked what varieties I had instead of adding the plants. Does he still do that?

Not from 2 September 2026: the plants go in first, and the variety question comes after. Tell him two raised beds, one with tomatoes and peppers, the other with celery, lettuce and spinach and he proposes the beds and all five plants in one go, then asks what kinds they are in the same reply.

A plant without a variety is a complete record. Its page shows the variety picker, so you can add the kind whenever you know it — the same way you would if you had added the plant by hand. Rafe used to hold back the plants whose variety he did not know and ask first; that was Rafe being more careful about a variety-less tomato than the app is.

Nothing is written without you. Each plant still arrives as a card you can check and undo, and the kinds question is a question, not a condition.

I told it I planted lettuce in the back bed, and instead of adding it the app asked whether that is the lettuce I already have. Why?

Because you might mean the one that is already there. From 2 September 2026, when Rafe goes to add a plant and that crop is already growing in that same bed, the app asks instead of guessing: You already have Lettuce in the back bed — is it the one you have, or another? There is a button for each plant it might be and one more, Add another Lettuce. Tap the one you have and nothing is added. Tap Add another and it goes in as a second plant.

Everything else you said in the same breath still goes in. New beds, new crops, plants in other beds — those are added straight away. Only the plant that might be a repeat waits for your answer.

⚠ It is the app asking, not Rafe. A plant in a different bed is never treated as a repeat, and neither is one you have finished with: the question only comes up when the same crop is growing in the same bed right now. If Rafe asks you the same thing in his own words instead, that is Rafe being careful before he adds anything — the card is the app catching the case where he did not.

Rafe said "let me add those for you" and nothing was added. Is that caught too?

It is now, from 3 September 2026 — and it was not before, because of one missing word.

There is already a guard that catches Rafe promising to write something down when nothing was written. ⚠ It was looking for a promise that starts with "I" — I'll add that, I will record it. Testing six ordinary sentences about planting kale, Rafe said "Let me add those for you" — no I anywhere — and the guard let it straight through. Nothing was written and nothing said so.

Two more phrasings slipped through the same way: "I'm going to record that" (the guard knew I am going to but not the contraction), and any claim that he had already done it.

And the sentence you get instead is more careful now. It used to say "that is not a crop I hold" — which was true for the case that first prompted it and wrong for a plant we do stock. Rafe cannot tell those two apart at that moment, so it no longer guesses why. It says nothing has gone into your garden, tells you saying it again will add it, and mentions "add it anyway" as something to use if it told you it does not have that plant — an option, not a diagnosis.

It said it would write something down. How do I know it actually did?

Because from 31 August 2026 it tells you when it did not, in its own line, above the answer.

Rafe writes the reply before the app has decided whether the write is allowed — so the reply can say "Got it, I'll write that down" about something that then does not happen. We found exactly that on 31 August, testing against a garden with three pole beans in it: "I watered the beans-pole" could have meant any of the three, so nothing was written, which is right. And the sentence said it had been.

⚠ That is the worst thing this app can do. Not a wrong answer — a wrong reassurance about your own records, which you have no reason to go and check.

So the app now says so itself: Not written down — I could not tell which plant you meant. It is the app speaking, not Rafe, and it appears whenever something was wanted and did not happen, whatever the paragraph above it says. The line that tells you something WAS written has worked this way for a while — it is built from what actually reached your records rather than from what was allowed. This is the other half of it.

When it cannot tell which plant you meant, say which one and it goes straight in.

If I tell it why I want a reminder, does that make it more urgent?

No, and that is deliberate. Say "remind me to water the tomatoes Thursday, I'm away for the weekend" and your reason is saved with the job and shown when Thursday comes — so you see why you asked instead of a bare instruction you have to work out again. What it does not do is push the job up the list. An urgency someone set by hand goes stale, and then it sits above a real frost warning, and then nobody trusts the top of the list at all.

The line saying what Rafe did just said "noted" twice and I could not tell what changed. Is that fixed?

Yes, from 8 September 2026, and it is a list now rather than a sentence.

It used to run everything together: Already done — noted Front Porch; noted Chinese (Mormon) Apricot; made a garden zone called backyard; noted Chinese (Mormon) Apricot. Two of those were corrections to your records and both came out as the word "noted", which says nothing about what is now different.

Now each thing gets its own line, with its own name first and what is now true of it beside it. A new growing space says new garden zone on its own line, so anything put into it on the next line reads plainly. A correction says which part changed — updated the kind it is, updated its name and how much sun it gets — in those words.

Why not old value and new value side by side? Because half of what a reply does is bring something into existence, and a brand new bed has no old value to show. A pair like that would have to print a dash or the word "none" on the left of every created thing, which reads as a mistake. Naming the thing and saying what is now true of it works for both.

Nothing about undo changed. Undo all of this and Change or undo are where they were, at the end of the list.

Rafe did something in the chat and I want it back. Where do I go?

There is a link right there in the line that says what he did — new on 31 August 2026.

When Rafe writes something down or makes something, the chat says so: Already done — added 6 Sungold; watered. That line now ends with Change or undo, which takes you to the list of everything he has done, with a way to reverse each one.

That screen has existed for a while and there was no way to reach it from the chat — which is the one place you would think to look, because it is where you just watched it happen.

When Rafe has not done anything, there is no line and no link. A quiet turn looks quiet.

And from 2 September 2026 there are two ways back, side by side. Undo all of this reverses everything that one answer did — the zone, the beds, the plants — in one tap, right there under the answer, and the line stays with each thing crossed out so you can still read what happened. Change or undo still takes you to the full list when you want to keep six of the seven. ⚠ A bed that already has a plant in it that Rafe did not make stays, and the line says so rather than pretending it went.

Rafe added several plants and the line under the answer just said "added a plant" for each. Which is which?

Fixed 2 September 2026. The line now names each one — Already done — made a bed called Back bed; added Tomato; added Bell Pepper — and the What Rafe did list shows the same names beside each row.

The cause was on our side. When Rafe adds a plant you have not named, the app names it for you, after the variety if you gave one and otherwise after the crop, numbered if you have more than one — Tomato, Tomato 2. ⚠ That naming was only happening when you added a plant yourself. A plant Rafe added on his own had no name at all, so the line had nothing to say and the Garden screen showed a blank where the name goes.

Now the plant is named the moment it is made, whichever of you made it. You can still rename it from its own page.

If you undid those plants one by one from the list, that is why it was hard to tell them apart — the rows had no names to show. Undoing everything from one answer in one tap is coming next.

I told Rafe my beds are in my main garden, and he named them "main garden raised bed 1" and "main garden raised bed 2". Why?

Fixed 2 September 2026 — Rafe can now make a garden zone.

A garden zone is the place a bed sits in: main garden, back yard, the community garden plot, the greenhouse. Beds go inside a zone; plants go inside a bed. Until today Rafe could make beds and add plants but had no way to make the zone — so when you said “in my main garden” he did the only thing he could and put the place into each bed’s name.

Now he makes the zone, puts the beds inside it, and the line under the answer says so: made a garden zone called main garden; made a bed called raised bed 1; made a bed called raised bed 2; added Tomato… The zone shows up under Garden zones at the bottom of your Garden screen, and on the What Rafe did list like everything else Rafe makes — with an undo, once the beds inside it have gone.

If you already have a zone by that name, he uses it rather than making a second one. And a bed you do not put in any zone is still fine on its own — most gardens never need one.

⚠ Two beds Rafe made before today with the zone in their names can be renamed from each bed’s own page, and grouped under a zone from the Garden screen.

Rafe said "Got it — it is kept", and the line under it said it was not written down. Which is right?

The line under the answer, always — and from 2 September 2026 Rafe no longer says either.

Rafe writes his answer before the app has decided whether anything landed in your records — so any sentence from Rafe about what was recorded was a guess, and twice in three tries he guessed wrong about a plant you did not have. The line under the answer is built from what actually reached your records, so it cannot be wrong in that way. Rafe now says the gardening part and leaves the record-keeping sentence out.

The order changed too. That line — Already done … or I did not write that down — used to sit above the answer, which read as finished-then-starting: Already done followed by I’ll add both beds. It now comes after the answer, so the app’s word is the last word.

I mentioned a bed I already have and now there are two of them with the same name.

Fixed on 2 September 2026. If you said something like "I put lettuce in Raised bed 2" about a bed you already had, the app could make a second bed with that same name instead of using the one you meant. Nothing on screen told you.

And it got worse quietly. Once two beds share a name, the app cannot tell which one you mean — so it asks which bed?, and the two answers it can offer read as exactly the same words. Every sentence naming that bed failed from then on, with no way back through the chat.

Now a bed you already have is used, never made twice. Say a name your garden already holds and the plant goes into the bed that exists. Two mentions of one new name in a single sentence also make one bed, not two.

If you already have a pair, open either from your garden, and rename or delete one — a bed’s own page has both. Deleting a bed also deletes the plants in it, and it tells you how many first, so rename is usually the safer of the two.

Rafe wrote a pest on my plant. Does he say what he saw?

From 4 September 2026, yes. When Rafe records something he judged — a pest, a disease, a plant that looks hungry, or something he cannot settle — he now writes down what he saw in plain words beside the entry: holes eaten in the lower leaves, clustered on the underside of the third leaf. Not the name again, and not advice — the thing that made him say so.

That note is for Rafe as much as for you. Rafe reads your journal every time you ask him something, and the evidence reads back with the entry, so next month he can tell whether the same thing is happening again rather than only that a word was written down. Photo diagnoses have carried this since June; typed conversations did not, so a pest you described in words was logged with its name and nothing else.

Your own words stay yours. If you retype the note before saving, nothing of Rafe's is attached to it. And the entry now remembers which conversation wrote it, so a puzzling line in your journal can be traced back to what was said.

Can Rafe find something from months ago, like the last time I had mildew?

From 4 September 2026, yes — he searches the whole journal. Before this Rafe was handed your last twelve entries and could look up one plant by name, so have I had aphids on the kale before? was answered from whatever was in the last twelve, which read as no once the aphids were the thirteenth entry back.

Now Rafe is told how many older entries exist and can search all of them by topic — pests, disease, feeding, watering, weather, harvest, pruning — or by a word, across every plant and every season, and read what he saw at the time. Ask when did I last feed the tomatoes or has the squash had mildew before and he looks before he answers. If nothing matches, he says so rather than pretending to remember.

More on this

• Asking about a photo
• Hardening off
• Your garden dates
• Photos