Mohammed Ali Chherawalla

Posted on Oct 27 Originally published at getoffgridai.co

How to Retry, Edit and Queue Questions to Local AI on Your iPhone in 2026 (Every Answer Kept)

#ios #ai #privacy #llm

You asked the model on your iPhone a question and the answer was close, but not right. You want another try, and you want to keep the first one. Or the model is still writing and you already know what you want to ask next. Off Grid AI Mobile (OGAM) handles both. Every retry and every edit keeps the old answer as a version you can go back to, and anything you send while a reply runs waits in a queue until the model is free. The model runs on your iPhone's own chip, so none of this leaves the phone.

What you need

An iPhone on iOS 17 or later with a chat model downloaded. Any GGUF chat model works, for example Qwen 3.5 2B or Gemma 4 E2B. If you have not set one up yet, start with How to Run Local AI on Your iPhone in 2026.
Off Grid 0.0.117 or later from the App Store. Versions and the queue arrived in the 0.0.115 beta and are in 0.0.117, the current App Store release. Download OGAM for iPhone.
No account and no payment. Both features are in the free app.

How the message menu opens

Every message in a chat has a menu. Press and hold the message bubble, or tap the three dots (...) next to its time, and a sheet called Actions slides up. The menu does not open on a reply that is still being written.

What the sheet offers depends on whose message it is:

On your own message: Copy, Edit and Resend.
On the model's answer: Copy, Edit and Regenerate, plus Speak and Generate Image when they apply.

Step 1: Get a new answer and keep the old one

Press and hold the answer you want to replace, or tap its three dots.
Tap Regenerate.
The chat ends at your question while the model writes a new answer.

When the new answer is done, a small switcher appears under it: a left arrow, a count such as 2/2, and a right arrow. Tap the left arrow to go back to the first answer. Tap the right arrow to return to the new one.

Resend on your own question does the same thing from the other side. It runs the question again and adds the reply as a new version beside the old answer. If your question asked for an image, Resend draws an image again instead of answering in text.

If a retry fails before it saves an answer, the chat goes back to how it was. Nothing is deleted.

Step 2: Fix your question and ask again

Press and hold your question, or tap its three dots.
Tap Edit. The EDIT MESSAGE sheet opens with your text in it and the keyboard up.
Change the text and tap SAVE & RESEND. Tap CANCEL to leave it as it was.

The edited question becomes a new version beside the original, and the model answers it. The original question keeps its own answer and every turn you had after it. The switcher appears under your question this time. Step back to the first version and the whole conversation that followed it comes back with it.

Step 3: Correct an answer by hand

Sometimes the answer is nearly right and you only want to fix a number or a name before you build on it.

Press and hold the answer and tap Edit.
Change the text and tap SAVE.

On the model's answers the button reads SAVE, not SAVE & RESEND, because nothing runs again. Your corrected text is saved as a new version of the answer, and the model's original stays one tap away in the switcher.

Which version the model reads

The model reads the chat as you see it: the versions you have selected, and nothing from the ones you stepped away from. To continue from an earlier answer, step back to it with the switcher, then type.

Step 4: Queue your next questions while the model writes

A long answer from a 4B model on an iPhone takes a while. You do not have to wait to type.

While a reply is running, type your next message.
Tap the send button. The message does not interrupt the model. It goes into the queue, which shows above the text box with a count such as 1 queued and one row per message.
Keep typing and sending. Each message joins the queue.

When the reply ends, everything in the queue for that chat goes to the model together as your next turn. The queue belongs to the chat you wrote it in.

Each queued row has two buttons:

The return arrow sends that one message now. The running reply stops and the message goes on its own.
The trash icon removes that message from the queue.

The x on the right of the queued badge clears the whole queue.

To skip the queue for a message you have not sent yet, press and hold the send button instead of tapping it. The reply stops and your message goes at once.

Step 5: Stop without losing what you queued

When the text box is empty, the send button turns into a square Stop button. Tap it and the model stops writing.

A Stop does not throw your queue away, and it does not send it either. The queue holds and the badge reads Paused after Stop. Tap Send next to it when you are ready, or remove the messages you no longer need first. If a reply fails, the queue holds the same way and reads Paused after an error.

Switching apps mid-reply on iPhone

iOS does not let an app use the iPhone's GPU while it is in the background, and Off Grid runs the model on the GPU through Metal. So if you swipe up to another app while a reply is running, the reply stops where it is. Before 0.0.117, the model then answered nothing until you ejected it and loaded it again.

From 0.0.117, the partial reply stays in the chat with a note under it: Reply stopped early. Retry to continue. Your next message, or a Regenerate on that answer, reloads the model by itself and runs normally. The retry becomes a new version, so the partial answer is kept too.

What this does not do

A reply does not keep running while Off Grid is in the background. Stay in the app until a long answer finishes, or retry it when you come back.
There is no list of all versions at once. You step through them one at a time with the arrows.
The queue does not send messages one by one. Everything queued in a chat goes as a single turn when the reply ends. Use the return arrow on a row to send one message by itself.
Regenerate uses the model and settings you have now. For a different kind of answer, switch model or change its settings first, then tap Regenerate.

FAQ

Can I regenerate an answer in a local AI app on iPhone without losing the old one?
Yes. In Off Grid, press and hold the answer and tap Regenerate. The old answer stays as a version, and the arrows under the new answer switch between them.

What happens to the rest of the chat when I edit an earlier question?
It is kept. The edited question is a new version, and the original keeps its answer and every turn after it. Step back with the arrows and that thread comes back.

Can I type while the AI is still answering on my iPhone?
Yes. Messages you send while a reply runs wait in a queue above the text box and go to the model together when the reply ends. Press and hold the send button to stop the reply and send at once.

Why did my reply stop when I switched apps?
iOS pauses GPU work for apps in the background, so the local model cannot keep writing. Off Grid keeps the partial answer and marks it Reply stopped early. Tap Regenerate or send your next message and the model reloads by itself.

Does any of this need internet?
No. The model, the versions and the queue all live on your iPhone. Airplane mode works.

DEV Community