Long replies are where AI is most useful and most annoying: a deep research run or a big refactor can take minutes, and the browser gives you no signal when it lands. Neither chatgpt.com nor claude.ai fires a desktop notification when a response completes. You either stare at the stream or you forget the tab entirely.
The workarounds people actually use
- Pin the tab and watch the favicon. Some UIs subtly animate the tab title while generating. Unreliable, provider-specific, and you still have to look.
- Split-screen the chat. Works for one session. The whole point of parallel AI is that you have five.
- Sound cues from the OS. Some voices-of-experience mute everything except the chat tab and listen for the reply chime — where a provider has one. Most don't.
- Ask the model to say something searchable. Ending prompts with “finish with the word DONE” helps you scan, not get notified.
What an actual solution looks like
The state you care about is readable from each tab's page — the stop button, the streaming markers, the composer state. AI Tab Switch reads exactly that, locally, across 29 providers including ChatGPT, Claude, Gemini, Perplexity, DeepSeek and Copilot, and turns it into three kinds of signal:
- a desktop notification when a reply completes — and a distinct one when a tab becomes blocked (asked you a question, wants a permission, hit a rate limit, logged you out), because blocked states are the expensive ones;
- a toolbar badge with the count of finished sessions, for when notifications are muted;
- an optional chime, plus batch mode — “only notify me once three tabs are done” — so notifications never become their own noise problem.
Everything is computed in your browser; page content never reaches a server (the privacy policy spells that out). If a completion signal is all you wanted, that's the whole setup: install, reload your AI tabs, keep working until the notification comes to you.