Two small settings that change how the app behaves around a message: one tells you when a slow answer has arrived, the other warns you before you send something you might not mean to. Both are among the features of Secret Chat AI, the AI aggregator.
"Notify me when the answer is ready"
Deep research and reasoning-heavy answers can take a while. With this on, a system notification appears the moment the model replies, so you can switch to another tab and get on with something else instead of watching a spinner.
- It only fires while the app is in the background — it will not interrupt you while you are looking at the conversation.
- It never contains your chat. The notification says your answer is ready and nothing more, so a message on a locked screen or a shared display gives nothing away.
- Turning it on asks your browser for notification permission. If notifications are blocked for the site, or the browser does not support them, the toggle says so instead of failing silently.
Personal-data (PII) warnings
With this on, the app scans a message before it is sent and — if it looks like it contains personal information — shows you what it found and asks whether you really want to send it. You can send anyway or cancel and edit. The check runs in your browser, on your text, before anything leaves your device.
This matters because it addresses the one risk the rest of the product cannot: Secret Chat AI removes you from your queries, but it does not remove the data from your messages. A prompt with a name, an account number or a diagnosis in it is still a prompt with those things in it. The warning is a prompt to think, at the only moment when thinking still helps.
Its limits, stated plainly
This is a heuristic, not a guarantee, and it is off-by-default territory for good reason:
- It misses things. Pattern-matching cannot recognise every identifier in every format, and it has no idea that a sentence about "my brother's clinic in Ann Arbor" identifies someone.
- It over-fires. Expect false positives — reference numbers and ordinary strings of digits that happen to look like something sensitive.
- It is tuned mainly for US formats. Its coverage of other countries' identifier formats is weaker, which is worth knowing if you are not in the United States.
- You can turn it off in Settings if the false positives outweigh the value — and many users reasonably do.
Treat a clean pass as "nothing obvious was spotted", never as "this message is safe to send". The judgement about what a model needs to see is yours, and there is no setting that can take it over. If a detail is not necessary for the answer you want, the strongest control available is simply not typing it.