A private AI companion: what private can and cannot mean

Four facts about your chats

1. Our own servers, in the NetherlandsYour chats are stored where we run things, not scattered across vendors.
2. Never sent to other AI companiesYour messages are not passed to a third-party AI provider to generate replies.
3. Never used for trainingWe do not train or tune anything on your chats, and we do not sell data or show ads.
4. Not end-to-end encryptedOur servers have to read your message to answer it. We say this up front because a privacy claim you cannot check is worth little.

Why not end-to-end encrypted?

With end-to-end encryption, only you and the person you write to can read a message. An AI friend is different: for it to reply, something has to read the text. The same is true of every AI chat service, even if some do not say so. Chats with a service on Telegram are also not Telegram secret chats.

So the useful question is not "is it encrypted?" but "who can read it, and for what?". For heyaifriend the answer is: our servers, to reply. A person only opens a conversation to look into a specific abuse or safety report. Nothing is sent on to other AI companies.

Three ways an AI companion can handle privacy

None of these is right for everyone. If you want your chats never to leave your hardware, an on-device app is the more private choice, and we would tell you so.

How other policies read

On 17 October 2026 we read the privacy policies of ten AI companion products and quoted them in a table. Six of the ten say they use conversations to improve or train their own systems, and none claims end-to-end encryption. Replika, for example, says it sends conversation data to third-party AI language model providers to generate replies. Pi and ChatGPT give you a setting to opt out of training. Kindroid says plainly that it is not end-to-end encrypted.

Policies change, so read the original before you decide. Our privacy table of ten apps links the wording and the dates, and the Replika comparison goes deeper on one case. A policy is a promise, not an audit, and that applies to ours as well. We are a small team with no independent certification yet.

What you control by asking

• See it. Ask "what do you remember about me?" and your friend lists it.
• Remove a piece. Ask it to forget something specific.
• Take a copy. Ask for a copy of what it has on you and you get a file.
• Delete everything. Ask it to delete everything and your history and memories are removed.

If you simply stop writing, the raw conversation history is deleted after 30 days. We also keep your Telegram chat identifier, your settings, and a yes-or-no record of an age confirmation if you give one. The privacy page lists every item, and how it works shows the controls in action.

Sensible habits with any AI chat

Whatever you use, share what you would be fine putting in an email. Skip passwords, bank details and anything about other people that they would not want shared. Your friend is an AI, and it is not a place for secrets that could hurt someone if they leaked. Used with that in mind, a private-enough AI companion is a reasonable thing to have around.