Why is Dayflow exchanging data with websites?

[neural-lace]

Little Snitch Mini is showing the app exchanging data with a number of websites (these are all be ones that i have visited recently).

The Github, Dayflow, Tracking websites seem explainable, but why the others. The data is on the order of bytes and kilobytes, so quite small. But I don't understand why it's needed at all.

Here is an example screenshot:

[JerryZLiu]

Ah so that's a feature I decided to implement. We query websites for their favicon, ex) we do a get request on youtube.com/favicon.ico

It's not exchanging any information, just requesting the icon so we can show them on the timeline. Does this seem like a bad idea?

[mikehardy]

"It's not exchanging any information"

This is not correct, simply opening the connection and issuing the request generates quite a trail of information that is logged via ISPs, DNS providers, and the site itself. Much of which can be loosely associated with the originating IP along with whatever information you're sending as user-agent etc.

One of the reasons I thought Dayflow was neat is because it could be network-isolated / local data transfer only and it seemed philosophically that was the intent.

This violates that intent

[JerryZLiu]

Got it, that makes sense. I think I just assumed that people would have already interfaced with said website so it's not really leaking any information that wasn't already leaked. Will add something to let people toggle that off.

FWIW, it doesn't make these requests for people using the local only llm version.

[mikehardy]

That is a reasonable assumption given it is just favicon, but at the same time private is private :-)

Mostly replying in this thread that is otherwise pretty resolved via your comment to say that I appreciate your interaction with feedback, just from one developer to another. Everyone wants something and it all takes time to respond to, I know that, so thanks.

[neural-lace]

I can't say if it's a good idea or not, but what I can say is that people are likely going to be super privacy focused for this app since it see everything you do. If you keep it, then worth mentioning it explicitly in the readme.

For myself, if you offered a toggle to turn it off, I would turn it off.

[mikehardy]

This and the analytics/crash report should be default off, opt-in IMNSHO
Feel free to pop a dialog during onboarding that encourages enabling the features, but opt-in is user-privacy-respecting, any other stance that accesses network services is user-privacy-hostile (again, IMNSHO)

[quinncomendant]

You can use DuckDuckGo's favicon service to fetch icons instead of sending a request to every website. That's more privacy-preserving, efficient, and will look less suspicious.

You can see how KeePassXC implemented this.

[squatto]

I looked for some other favicon services, and there are a number of different options. Here are some:

• https://www.google.com/s2/favicons?domain=youtube.com
  • Optionally pass sz=# to request a size, e.g. https://www.google.com/s2/favicons?domain=youtube.com&sz=128
• https://favicon.yandex.net/favicon/youtube.com
• https://icons.duckduckgo.com/ip3/youtube.com.ico
• https://f4.allesedv.com/32/youtube.com
  • Replace 32 with a size up to 144
  • Docs: https://favicon.allesedv.com/

As an idea, to alleviate concerns about using additional services and exposing personal IPs/etc., you could run a single Dayflow-hosted service that takes a domain/size/etc., pulls the favicon from whichever third-party service will provide it, and serves it to the Dayflow app. Then there's only a single external service being hit by the Dayflow app.

[squatto]

It looks like you've already implemented pulling favicons from google.com. Excellent 👍🏻

https://github.com/JerryZLiu/Dayflow/blob/main/Dayflow/Dayflow/Core/Net/FaviconService.swift