Hacker News new | ask | show | jobs
Show HN: Semantic Search for Ctrl+F (chrome.google.com)
14 points by iamscout 1516 days ago
Hi HN!

Ctrl+F sucks because I can never seem to find what I'm looking for. I built an NLP-based alternative that works pretty damn well.

Let me know what you think.

P.s. it will soon support image searches

8 comments

Honestly I wouldn't even consider installing this because there is zero information about how it works and what it does with the data I type into it.

Are my search terms sent to some random server to be "processed" and/or logged for analytics purposes, or is everything done locally?

The single-line privacy policy is [woefully bare](https://sites.google.com/view/scoutcrx/privacy).

I looked at the source code and found `new a.OpenAIApi(new a.Configuration({ apiKey: ... }))`, so I assume it’s calling a remote endpoint.
Thanks for the feedback. I'm still busy putting together the parts of this project which is why the privacy policy is barren. Scout doesn't save what you search, it's doing a direct call to a search API, but at no point is what you search recorded.
Your claim does not line up with: https://openai.com/api/policies/privacy/

You're exposing your API key, plus I wonder if everyone trying your extension has made their IP address visible to you on the 'usage' page mentioned here? https://help.openai.com/en/articles/5112595-best-practices-f...

The exposed API key is temporary while a backend is setup. The IP address is not stored by Scout - will look into the OpenAI policy.
"Any data about your use of Scout is not knowingly sold or offered to third parties or stored on our servers."

Privacy policy seems to imply that data is being collected - would be good to clarify that. Does it rely on a server/API?

It relies on an external search API. No data is recorded or saved
Your claim does not line up with: https://openai.com/api/policies/privacy/
Will look into this.
not sure, but could the privacy policy just be a requirement of Google and no data collection is taking place?
I had a look at the code. Despite the important note on https://github.com/openai/openai-node it appears to be in use here, leaking the author's API key.
Yes this is the case.
This is not the first time that we've had semantic search as a "Ctrl-F" replacement in chrome:

I actually cited the precursor to this in one of my papers:

https://chrome.google.com/webstore/detail/fuzbal/lidjpicdkcg...

Also actually open source: https://github.com/ijkilchenko/Fuzbal (MIT)
Is this using OpenAI‘s GPT3 search endpoint? Someone here mentioned a code snippet from the extension that seems to suggest that. Reason I‘m asking: Last time I checked, OpenAI‘s GPT3 „search“ endpoint actually performed a lot worse than, say, a BM25 or fastText search (which has the added benefit that you can run it on your own server).
Good point, I'll look into this to see if performance is improved.
If you use Python on the backend, rank-bm25 is a great library for quick and dirty bm25 search.
Please do consider updating the image to be something that shows what is actually going on, because that one just makes it look like your extension draws highlights on a webpage. Bonus points for some animated gif or video or something showing how it changes as you type more stuff
Good idea, will get on that.
Cool idea. I want to be able to semantically search the contents of files on my computer and an archive of my bookmarks
It's just a browser extension for now - maybe one day soon.
Is this open source? Seems like it might be small enough to be great for learning from.
I'm still seeing where I will take this project, but yes I agree with you.
Where's the source code
You can download it with an extension like: https://chrome.google.com/webstore/detail/chrome-extension-s...
Where's the source code to that?
I know you are trying to be funny but you can just install it in a new profile and then inspect the extension on your machine, or use a website that downloads Chrome extensions or just use curl to download it from Google (https://stackoverflow.com/a/19665274).