Every message in a conversation carries its own set of actions. They are how you fix a badly-worded question, get rid of something you would rather not keep, or check exactly how an answer was handled. This is one of the features of Secret Chat AI, the AI aggregator.
What you can do with a message
- Delete — removes that single message from the conversation. See below; it is the action most other AI apps do not give you.
- Copy — puts the message text on your clipboard.
- Edit — reopens your own message prefilled, so you can reword it and send it again. It does not overwrite the original: the reworded version is added to the conversation as a new message, and the earlier one stays where it was until you delete it.
- Retry — asks again. Useful when an answer was truncated, went off on the wrong track, or when you want to see what a different model makes of exactly the same question.
- Thread — branches a new conversation from this point. See threads.
- Session Privacy Report — opens the PDF described below.
Deleting individual messages and images
This is the one worth knowing about, because the big AI apps do not offer it. In ChatGPT and Claude, deletion is all-or-nothing at the level of a whole conversation: you can throw away the entire chat, but you cannot reach into it and remove the one message you regret. So a single ill-judged paragraph in an otherwise useful three-hour conversation leaves you choosing between keeping it and destroying all the work around it.
Here, deletion is per message. Hover the message, press delete, confirm, and it is gone while everything around it stays. Generated and attached images are deleted individually too — each picture in a chat has its own delete control, so you do not have to remove the message to remove the image.
Two things this is genuinely good for:
- Cleaning up after a slip. You pasted something identifying, or something you would rather not have sitting in your history. Remove that message; keep the conversation.
- Keeping a long conversation useful. Dead ends, mistaken questions and images that did not come out can be cleared away, so what remains is the thread of work you actually want to reread.
Deleting is final — your history lives in your own browser, so there is no server-side copy for us to restore it from, and no trash to recover it out of. What deletion does not do is reach back to the model provider: the text reached them when you first sent it. That is what the report below is about.
The Session Privacy Report
Most AI apps ask you to take their privacy claims on trust. The report is our attempt to hand you evidence instead — and you do not have to request it. It is generated automatically for every answer. The report button on a message shows a spinner while it is being prepared and becomes active as soon as it is ready; you just wait a moment and open it. Inside is a PDF recording how that exchange was handled: which model actually answered, which provider it ran on, and — where the provider offers a per-response deletion API — that a deletion request was sent and what it returned.
Reports have been produced this way since 29 July 2026. Conversations from before that date have no report attached to them.
Downloading one gives you a file named after the provider response ids it covers, rather than an anonymous report.pdf. That is deliberate: it is the same identifier our own copy is filed under, so if you ever need to ask us whether a particular response was really deleted, the id in your filename is the one we search for.
Reading it, three outcomes are possible, and the report distinguishes them rather than blurring them together:
- Deleted. The provider stores responses and confirmed our deletion request for this one.
- Not stored at the provider. There is no stored response to delete — which is a stronger position than stored-and-then-deleted, not a weaker one.
- Per-request deletion is not needed or not offered. For DeepSeek USA, for instance, requests are routed only to providers that declare they do not store prompts, and that routing layer exposes no deletion API at all. The report says so explicitly rather than printing a misleading "not deleted".
A provider's own abuse-monitoring retention is a separate matter from the response object, and the report keeps those apart too.
What we keep of the report
To be straight about it: a copy of each generated report is archived in object storage on our side, so that a question months later about whether a deletion actually happened can be answered. What that copy contains is the report — model, provider, deletion status and response identifiers. It is metadata about the exchange, not the text of your prompt or the answer. Your conversation itself remains in your browser only; a prompt exists on our side only for as long as it takes to fetch your answer.
The limit worth understanding
Deleting a message in the app removes it from your device, and the report tells you what happened at the provider — but the two are separate acts. Deletion here does not travel back in time and un-send anything: the text reached the model provider when you first asked. Secret Chat AI is an anonymizer, not a content filter — it removes you from your queries, not the data from your messages. The most effective control remains what you choose to type in the first place; the app can warn you about some of it, but the judgement stays yours.