How to use AI without saving the conversation
- Incognito means the conversation is not written into your chat history.
- It pairs with local inference: the message was never uploaded, and now it isn't stored either.
- Useful for anything you'd rather not scroll past next month.
- It is a storage choice, not a privacy fix for cloud AI — that ship sailed before you typed.
Search engines learned decades ago that people ask machines things they'd never ask a person. Chat assistants inherited that, and then started keeping a permanent, searchable transcript of it.
What incognito does and doesn't do
It stops the conversation being persisted to your history. That is the whole promise, and it is worth being precise about — a mode that quietly kept a copy would be worse than not offering one.
What it does not do is make a cloud assistant private. With a hosted model your text has already left the building; a local toggle can't retrieve it. Incognito is only meaningful on top of inference that was already on your machine.
Using it
- Start an incognito conversation when you don't want it kept.
- Use it normally — same models, same quality.
- Close it and it isn't in your history.
The companion window honours the same setting, so a quick question there doesn't get recorded when the main window wouldn't have been.
When to reach for it
- Health, legal, financial questions — the classic "I'd rather this not be a permanent record" category.
- Anything about other people, which is their business as much as yours.
- Shared or work machines, where history is effectively a shared document.
- One-off scratch work that would only clutter your history.
What it protects, stated exactly
Incognito is a local-persistence setting, and it is worth being precise about the boundary because "private mode" means very different things in different apps.
It does two things. The conversation lives in an ephemeral store
that is never written to disk, and the turn is marked so the model skips memory
recall and memory saving. Nothing from the chat is read out of your memory
store, and nothing from it is written back — including from an attached document, so
dropping in a private file cannot quietly seed facts you will see referenced in a
conversation next week. Toggle it with ⌘⇧I or /incognito. A persona with
memory switched off behaves the same way without you having to remember the toggle.
It is not a network setting, and it does not unsend anything. This matters most if you have web research on — a Pro feature, off by default, opt-in per conversation. When it is on, your query goes to DuckDuckGo, Mojeek or Wikipedia, and incognito has no power over what those services logged. It governs your disk, not their servers. If a question is sensitive enough for incognito, it is sensitive enough to leave the web toggle off, and the two settings are independent on purpose.
The part people expect to have to check — whether the prompt itself went anywhere — is not an incognito question at all. The model runs on your Mac in every mode. Incognito changes what survives afterwards, not who saw it.
Common questions
Does incognito make AI private?
It stops the conversation being saved to your history. Privacy from the provider comes from running the model locally in the first place — incognito is the storage half of the question.
Is the conversation still processed on my machine?
Yes. Incognito changes whether the chat is stored, not where inference happens; that was already local.
Can I recover an incognito chat later?
No, and that is the point. If you need it, use a normal conversation.
Does the companion window respect it?
Yes. The small always-on-top window honours the same incognito setting.
Try it on your own Mac
One signed, notarized download. No account, no token bill, and it keeps working with the Wi-Fi off.
Get Outlier for Mac