Least privacy-invasive ways to interact with AI/LLMs? (excluding self-hosting)

What I am aware of:

  1. Self-hosted LLMs
  2. Brave’s Leo Chatbot is built into Brave Browser, It is based on the Mistral 8x7B model, it is privacy respecting, but very rate limited.

I’m curious what other “least-bad” options are available? I’m not hoping for perfect-privacy, something akin to the level of privacy gained by using newpipe or freetube to access youtube would be a reasonable compromise in my eyes. Obviously more privacy would be ideal, but not a hard requirement for me.

1 Like

Brave Leo with Llama 2 13b model. It has no rate limit and it’s free. The result is acceptable, but I think it’s still behind Gemini and ChatGPT (it could be my use case).


You can also use Microsoft Copilot without an account, on a vpn and in a privacy browser

It would give you GPT-4, and assuming you clear cookies and cache, and your browser is pretty good at fingerprint protection, each session you have will be separate from the last


I didn’t realize that that was an option. Thanks for pointing that out.

Good to know. I wasn’t aware of this. I’m looking into it now.

(Ironically I’m using Copilot to assist in checking the subdomains and third party domains it is connecting to, to see what can be safely blocked with uBO :smiley: )

Query: What is the domain bat.bing.com used for?
Chatbot: The domain bat.bing.com serves a specific purpose related to Bing Ads Conversion tracking. When a user clicks on a Bing-powered ad and visits the advertiser’s webpage, the Bing Ads UET script sends a beacon or signal to the Bat.Bing servers hosted at bat.bing.com.

Query: can I block bat.bing.com without breaking copilot.microsoft.com?
Chatbot: Certainly! Blocking bat.bing.com should not impact the functionality of copilot.microsoft.com.


Which is self-hosted.

You may he able to block tracking, but keep in mind whatever text you write will still be sent to their server, even if your identifiers are obscured

agree, it’s a tradeoff:/

1 Like

No, Brave hosts it.

1 Like

That might be what they meant. Self hosted by brave (i.e. in house)

1 Like

Sure, then ChatGPT and Google Bard are self-hosted as well :laughing:

In all seriousness, it is actually important to be clear, otherwise people could assume it is built in to the browser. If you make the individual choice to trust Brave (and their comparatively better privacy policy vs OpenAI/Google/M$), that’s fine, but there are no technical privacy features built in.


For all of these goes: do not put personal information or anything close to it in these models.

Besides I personally currently use many but by far the best option out is perplexity.ai which has no limits and works without account. It surely isn’t privacy friendly with what it scrapes from the internet and what you put in. Results are however pretty good.


They recently switched to Mixtral 8x7B as the new default.


I was testing Copilot last night (without a VPN) and it was working fine, today, I tried to use it with a VPN, and I am being asked to ‘sign in’ to use it.

Have you personally used Copilot with a VPN?

I use a residential iP, so it isn’t detected, I knew that Copilot blocks Tor, but I wasn’t aware it blocks VPNs😮‍💨

I’ve got things working now. I think it must’ve been just a blacklisted VPN IP address. I can connect via VPN now.

I wanna reshare my previous post about different models.

Personally, I use different bots for different purposes such as personal, work etc. If you have a good computer, you can even use local models.

I prefer perplexity and hugging chat over ChatGPT and other models. Don’t forget to deactivate the option of the chatbot’s training on your data.

For work purposes, I use gpt-4 most of the time, because others may produce poor results, costing time on my work.


Perplexity seems interesting.

I’m curious which model you choose and if you are using the free or paid offerings? It seems like they have a handful of different product offerings.


I do see the ownership is:

  • Meta
    • Meta
      • Perplexity AI

Which is not great. Though perplexingly LLMs seem to be the rare area where Meta is not one of the worst actors in the space (yet). And Perplexity works without an account and behind a VPN so that is a plus.

[Edit: I can no longer find the source that originally gave me the impression Perplexity was a Meta backed venture. And I can’t find any other source that states that. So I believe I was wrong.

I think Phind is what you are looking for. It is free, don’t need account. You can simply clean cookies each time

1 Like

I am not sure whether Meta’s really the owner of Perplexity. AFAIK, it got most of its funds from Nvidia and Jeff Bezos (Amazon). It’s built on Meta’s open source Lllama 2 model, though, which is not the same as having Meta as its ownership, hence having a say in the project direction.

1 Like

I think you may be right, I may be mistaken about Meta involvement (looking into that right now). I recall reading that Perplexity was backed by Meta (as in funded by, not just using its open source model), but I’m struggling to find where I read this (I had thought it came from the Perplexity website).

Also, I think that the main Perplexity free offering does not actually use LLama 2, I believe the free offering is based on GPT 3.5, you can go to Perplexities “playground” to experiment with a number of different models though (including Llama 2 and many more.

edit: yeah I think I was mistaken, the only Meta connection I can find now is that one of the first investors in Perplexity was Meta’s Chief Scientist but that is just one of about a dozen early investors.