The Editor Stops Freezing Mid-Sentence, AI Studio Confirms Before Overwriting Your Work, and Snapshots You Can Finally Reach -- Plotiar Product Updates (August 2026)
A writer who splits her afternoons between a phone on the train and a laptop at her desk wrote in twice in one week. The first time, the app had frozen solid mid-paragraph and she'd had to force-close it just to keep typing. The second time was worse: she picked her laptop back up after a session on her phone, pressed Backspace once, and watched a character vanish from a word she hadn't touched. Both reports pointed at the same layer of the editor -- the part that reconciles what you're typing right now with what just arrived from somewhere else. This stretch went after that layer directly, and after a run of separate reliability problems in the AI tools that had been quietly costing people credits for work they never got to use. Here's what changed.
Two Devices, One Document, No More Freezing
Keep the same document open on your phone and your laptop and edit both at once -- which is exactly what autosaving, always-syncing software is for -- and the editor used to visibly seize up while it absorbed what the other device had just written: several seconds of the interface not responding to a keystroke, worse the longer the document. The editor now folds an incoming change in as an edit to just the part of the page that actually changed, instead of redrawing the whole document every time, and repaints only the area that moved. A stretch that used to freeze for several seconds now drops to a fraction of that, and a document sitting idle with a blinking cursor no longer burns time in the background at all.
The bug that deleted the wrong character came from that same reconciliation working off stale coordinates. When your other device edits text earlier in a paragraph you're both writing in, your cursor has to shift to stay on the same letter, and in a few specific situations it wasn't, so a single Backspace could land twenty characters from where you were actually looking. Undo had a matching problem: pressing Ctrl+Z right after a remote edit landed nearby could replay against text that had since moved, cutting a chunk out of something you never wrote yourself. Both are fixed -- undo now recognizes when a remote change has made its own memory of "what to reverse" unreliable and simply does nothing rather than guess, and the cursor now tracks the actual character it's sitting on through any edit elsewhere in the same paragraph. Redo got a smaller, related fix: pressing Ctrl+Y right after an undo now puts your cursor back exactly where the redone edit left it, not a step behind.
Typing Near A Footnote No Longer Breaks It
Footnotes are new enough that the rough edges were still showing. Add one at the end of a sentence and keep typing, and the editor could swallow everything you typed next into the footnote marker itself -- invisible, because a marker only ever displays its reference number regardless of what text actually sits inside it. Depending on where your cursor landed, this showed up as text vanishing right after a footnote, the cursor refusing to move past one with the arrow keys, or Backspace appearing to count invisible characters. Pasting near a footnote had its own version of the same bug, and so did typing in a language that builds a character over several keystrokes before committing it.
All of it comes down to one rule the editor now actually enforces: a footnote marker holds exactly its reference number and nothing else, so any text landing near one is redirected onto a normal, visible run beside it instead. The same fix also cleaned up three smaller bugs inside table cells that had nothing to do with footnotes -- the arrow keys skipping a beat at formatting boundaries, a selection spanning two differently formatted pieces of text in one cell not actually deleting when you typed over it, and, with track changes on, a keystroke in a cell occasionally landing one character off from where you'd typed it.
The Snapshots Panel Now Shows Your Snapshots
Every time an AI Studio generation is about to replace what's in a document, Plotiar quietly saves a copy first, and the Snapshots tab is supposed to be where you go to get that copy back. It wasn't showing them. The tab opened to a different kind of history built from your own ordinary editing activity, while the actual list of saved snapshots -- restore, rename, and compare included -- existed in the app but was never connected to anything a person could click on. It's connected now, on its own tab, with the other view relabeled "Version history" so the two stop being confused for each other. If you've ever been told to roll a document back after a generation and found nothing there, this is why, and it's fixed. The confirmation prompt guarding the restore step, and the panel itself, are also now fully translated into every language Plotiar supports.
AI Studio Asks Before It Replaces Your Work
The panel on the right side of AI Studio always works the same way: whatever comes back replaces the document you have open, including anything you wrote yourself before running it. The confirmation dialog only appeared on a second or later generation, which meant the very first run -- the one most likely to be overwriting your own original text rather than a previous AI result -- happened with no prompt at all. Every generation now confirms first, names the actual document, and says plainly that your own writing is included in what's about to be replaced.
On the billing side: if a generation comes back empty, or cut short with nothing usable in it, Plotiar no longer writes it into your document or charges you a credit for it. If it's cut short but still produced usable content, you keep that part and get a plain notice that it's partial rather than a silent gap. The same logic now covers AI project creation too, which previously could -- rarely, mostly on longer non-English prompts -- come back with a title and description but no actual content, and still charge for it. That can't happen anymore: if the full result isn't buildable, you get a real starter project instead of an empty shell.
Inspect, Analyze, And Lore Extraction Finish The Job
Word Inspect -- the tool that returns a full breakdown of a word or phrase, definition through alternative phrasings -- was quietly stopping partway through every time it ran, because the space it was given to answer in wasn't enough for everything it's asked to return. The fields near the end of the list, like metaphors and alternative phrasings, were never coming back. It's given enough room to finish now, looking up the same phrase twice is instant and free, and that no longer lands a duplicate charge on whoever asks next. Analyze had a quieter version of the same problem: a section cut short by its own internal limit could still pass its "did anything come back" check and get saved as finished with fields silently missing. A section like that is now marked partial instead of looking complete.
Lore extraction -- the tool that reads a manuscript and suggests entries for your lore book -- had two separate ways of losing results you'd already paid for. A routine background sign-in refresh could wipe suggestions that existed only on screen, with no way back to the finished job, and every way of closing the dialog discarded unsaved suggestions with no warning, next to a button labelled "Done" when it meant "discard." Results now survive a refresh, closing while something is unresolved asks first, and the button that adds everything says exactly that. Running the same document again while you're still deciding what to do with the first set of results is free now too, instead of billing you again for identical work.
A few smaller fixes rounded out the pass: a confirmation dialog opened from inside another panel could render invisibly behind it, which looked like the button had stopped working, and the footnote numbering control in the sidebar could keep showing "1, 2, 3" after you'd switched a document to Roman numerals, even though the document itself had switched correctly.
The writer with the phone-and-laptop routine wrote back after this shipped, this time about a chapter, not a bug.
Try Plotiar at app.plotiar.com.