[trees101]

it would be handy in the age of AI, to be able to dynamically scrub data that gets copied/pasted into the AI.

It's too easy to leak secrets, or even doxx yourself through file paths containing your name etc.

I'd love to find a tool that made scrubbing that data easy

[richbell]

This is sort-of describing https://docs.private-ai.com/webdemo/

[trees101]

thanks, that is close!

[nonamepcbrand1]

As a security engineer, I started building this tool but my feedback was so poor that no one cared pasting secrets, personal data into LLM chats.

[lexokoh]

We are working on this for the next release happy to get your feedback on it if it's possible

[trees101]

Great. I implemented my own simple prototype, a python script that edits my clipboard. I used the pyperclip module and a yaml file with a list of key words to substitute. Substitution is necessary rather than removal, so that the AI's responses are still useful.

I got basic functions working but there are some nice-to-have things missing.

E.g. bidirectional info preservation. Ideally if i change /my_full_name/ file path, I want it to be translated to /john_doe/ and when the LLM gives back its response, I want to be able to paste /my_full_name/ back.

Also, preferably it would be highly automated, where I have to manually run my script to edit my clipboard. Also, nice to have it work for non manual cases such as when using aider-chat.

Further down the line, automated redaction of screenshots.

[lexokoh]

This is great. I am happy to collaborate. If you fill out the contact form on the site, I'll contact you to try it out. Or open an issue.