Claude 'lost' hours of my chat history overnight, and I found it hiding on an invisible branch

Published Jul 31, 2026, 6:00 AM EDT Maker, meme-r, and unabashed geek, Joe has been writing about technology since starting his career in 2018 at KnowTechie. He's covered everything from Apple to apps and crowdfunding and loves getting to the bottom of complicated topics. In that time, he's also written for SlashGear and numerous corporate clients before finding his home at XDA in the spring of 2023. He was the kid who took apart every toy to see how it worked, even if it didn't exactly go back together afterward. That's given him a solid background for explaining how complex systems work together, and he promises he's gotten better at the putting things back together stage since then. I might have more devices running Claude than the average user, but I’ve also got local LLMs on most of those to test against. But whichever AI tool I’m using, I rely on memory and chat history to keep my home lab and experiments in order. With local LLMs, I expect to build the memory layer, but Claude has built-in memory and a chat transcript that is synced via the web version. You can’t trust agentic workflows lacking reliable logs, and I’ve been impressed with how Anthropic handles them. Until the morning that I found hours of results and benchmarking missing from chat. My iPhone edited a message as I was asleep This is a good example of why you should put your device away at night The night before, I was bouncing between devices, benchmarking FSR on the Steam Deck and ROG Ally X. Some of the chat was to troubleshoot FSR4 not showing up in certain games, while other parts were to record the benchmark results with my phone’s camera. I know I could do that manually, but it takes forever to record every part of the benchmark, and OCR does it in seconds. Then I went to bed, but checked through the conversation to see what the results showed and any other interesting bits of info. It seems I accidentally long-pressed a message from earlier in the conversation as I fell asleep, and tapped some random letters into the chat window. When iOS suspended the app, the pending draft message got committed as an actual edit. I didn’t press send on the message, but that didn’t matter to the app. Editing an old message doesn’t issue a correction; it forks the conversation from that point and the old conversation content gets hidden from view. Your chat history is a tree, not a log It is closer to code than conversations The issue I’d unintentionally triggered is an underlying feature of how most AI chat apps work. They’re not a flat transcript as you’d get from a phone conversation. They’re closer to a Git worktree, with branching lines of thought. The problem isn’t the design; it's the UI over top. Nothing in the interface tells you a conversation has multiple branches. The only indicator is a toggle that looks like under one message, which is tiny and doesn’t show up on every client. If it did, this wouldn’t be a problem. I would have woken up, seen the indicator about a forked conversation, and gone back to the first chat results. The other thing is that it gets worse the more clients you use. With multiple devices running Claude, the likelihood that one chat contains stale data increases. And it only takes one piece of old data to be synced to put you into a new branch with what appears to be missing data. The desktop app hid the evidence until I looked on the web version This was either a bug or bad design, but I reported it as an issue Once I figured out that the conversation had branched, not disappeared, I went looking for it. The pager arrows that Claude’s documents mention didn’t appear on the desktop app of my Windows PC. I browsed through the whole thing twice and moved my cursor over every message to show the retry, edit, and copy controls, where the branch indicator should be. Then I opened the same conversation on claude.ai in a browser, and the branch control was sitting right there. I clicked to switch to the other branch, and all the missing messages came back. Fine, mystery solved. I moved back to the desktop app, and the arrows had appeared there as well. It seems opening the website view updated the app content and made it show both branches. I’d stumbled on two bugs: the iOS app saving a pending message when the app was suspended, and the desktop app not showing that a branch had been created. I reported both, not expecting them to be fixed quickly. I was wrong. It’s only been a few days since this happened, and when I sat down to reproduce the issue for screenshots, I couldn’t do so. The desktop app showed the branch pager straight away, and the iOS app shows a warning that editing a message will restart the conversation from that point. Anthropic fixed the issues I was drafting an angry article about, and now I’m only annoyed at myself. How to get your "missing" messages back and stop this happening again The recovery takes ten seconds once you know where to look If you notice a missing piece of a conversation, the first thing to (not) do is send any new messages. Every new message buries the fork deeper, making it harder to find. Then open a web browser and go to claude.ai. The browser fetches the fresh state, and it was the only place that found the fork for me. Scroll to where your messages disappear, and move your mouse cursor over the last message. The branching indicator didn’t show up for me otherwise. It will look like , and clicking the left arrow will show the first conversation. There is one more thing you can do to avoid confusion. When you switch devices, treat it as the handoff it is and refresh the conversation before sending a new message. Pull down on the mobile app, or reload the tab or window on desktop. And close the apps you’re not actively using. Silent forks break the multi-device model but invisible ones break your trust I’m not arguing against branching here. It’s the best way to handle AI conversations, which might get edited at any point, and you want an accurate record of what happened. The problem was that some of the apps didn't show that a fork had been created, and I thought I’d lost hours of work. Assume this is how every AI chat app works, and you’ll know how to find forked data. Claude is Anthropic's family of frontier AI models, wrapped in a handy chat interface.