AI chat that doesn't store your conversations: what to check before you trust one

AI chat that doesn’t store your conversations: what to check before you trust one

2026-11-16

Using AI chatbot - artistic impression.

Using AI chatbot – artistic impression.

Plenty of AI chat services boast that they ‘don’t store your conversations’. This is a vague claim. It can be broken down into three elements: what the chat service stores, what the AI model provider (the company that supplies the AI model used by the chat service) stores and what information is available to link any of this data back to you.

Layer one: the chat service

By default, most AI assistants will store your conversation history on their servers, linked to your account. (Even in ‘temporary chat’ modes they will often do this for some time.)

Start by checking if the service stores any of your conversation history. If they do, how long is it kept for?

Take Obscurify.ai for example. They don’t store any of your prompts or the model’s responses on their servers. In fact, they are only held in memory (for performance reasons) for a maximum of 1 hour, and then permanently deleted. Your chat history is stored in your own web-browser (so technically anyone with access to your computer can read it, but you can also just delete it).

Layer two: the model provider

Usually the service you talk to uses models from a third-party provider. That is, your data (prompts and responses) is sent on to another company. The provider might log the prompts and responses, or use them for training.

Find out what route (there could be many) the service takes to reach a model, what assurances you get from them (contractual or just a written policy) and if these apply to all models or just some.

Obscurify.ai currently tags the models for which none of the providers that might serve them will keep your prompts and answers (a “zero data retention” or ZDR route) and for these models it won’t fall back to a route that might. Other models are available without the ZDR tag. The promise for the ZDR routes is currently the providers’ published policies (not a separate agreement with them).

Obscurify.ai also removes identifying metadata from your prompts (such as your IP address) before passing them on to the LLM provider, so that while it gets the text, it doesn’t know where it came from. If a service doesn’t explicitly state that it does the same, don’t assume it does.

Layer three: what still identifies you

Even if the service doesn’t store any conversations, it may still be possible to identify you via your account, your payment method, or your IP address.

Account: New Obscurify.ai accounts are a random 16 digit number. There is no email address, phone number or even a password. Only a keyed hash of the 16 digit number is stored on the server, so if you lose the number it can’t be recovered. There is also a free guest chat which doesn’t require an account at all, and doesn’t set any cookies.

IP address: IP addresses are kept for up to 30 days, mainly to prevent abuse, and then deleted. There is also a Tor onion address.

Payment: Payments can be made by card (using Stripe) or by crypto (Bitcoin, Lightning, or Monero). Payments by default will top up your account directly (and thus will be linked to your account on their end), but there is an option to pay for “blind-signed” vouchers instead, which breaks this link (your browser “blinds” the voucher before it gets signed). Of course a payment by card will be visible to Stripe and your card company, and a payment by Bitcoin will be publicly recorded on the blockchain. And Obscurify.ai will know the IP address that you made the purchase from, and the time that you later redeemed the voucher.

How to verify any of this

If the server side code is closed source, you’re limited in what you can verify.

With Obscurify.ai at least you can verify that no third party scripts are loaded on the pages (just open up the “Network” tab in your browser’s developer tools, the only external script is from Stripe and only loads when you initiate a card payment). There’s also a signed warrant canary that is updated monthly. The web client is open source so you can read through what is sent when you start typing.

Obscurify also has a ‘Local Mode’, where small models are run entirely in the browser. Your data never leaves your device (well, the first time you use a model it needs to be downloaded). It’s often only a few tokens per second.

A short checklist

Are your chats being stored on the service’s servers?

Does the AI model provider retain any data? If not, is this guaranteed in a contract or is it just a ‘policy’? Does this apply to all models?

What information is required to set up an account?

Can a payment be traced back to you?

What can you verify that they do (note you can’t verify much when the server code is closed source)?

A service passes or fails based on whether you get a satisfactory answer to these questions. (Not based on whether they advertise that they “don’t store your conversations”.)

AI chat that doesn’t store conversations explains in detail how Obscurify.ai addresses these issues.

AI chat that doesn’t store your conversations: what to check before you trust one
We use cookies and other technologies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it..
Privacy policy