PISIGuard: Protect your personal and sensitive info when you chat with AI | Hacker News Reader

PISIGuard: Protect your personal and sensitive info when you chat with AI

github.com

21 pointsmohamed_am8314 comments

Interesting.

1. Suggestion: Personally I love when projects mention whether they’ve found other existing software that does something similar, and if so how this app compares.

2. On iOS when I click into the project it opens into the GitHub iOS app but there’s a big empty box below the big logo. Clicking the box opens the video in my browser. Maybe it’s a GitHub app issue? I didn’t spend any time further on that.

3. Minor but I couldn’t be sure what pisi stands for - the page seems to use “personal and sensitive information” (PASI?) but maybe it stands for “personal information sensitive information” even if the page doesn’t use that phrase?

Hey, thanks for the feedback!

1. Correct. Honestly I haven't found any (I'd use it if I did). After release I was told enterprises have smth similar under the category Data Loss Prevention (DLP).

2. It's on me the logo is huge, I made it smaller. Video display is better on browser (not desktop app). If you scroll past the big box (the video) you can find the rest of the documentation.

3. Spot on, I added the full name to README and made the logo smaller. The name btw is "Personally Identifiable & Sensitive Information Guard"

It's heavier and isn't designed as a browser extension, but in terms of existing software for anonymization there's stuff like https://presidio.dataprivacystack.org originally from Microsoft.

Some people are confused as to what this is, so I thought I'd bring up this sentence from the Readme:

> PISIGuard runs entirely in your browser. It spots names, email addresses, phone numbers, credit card numbers, passwords, API keys, and more, replaces them with safe placeholders, then puts the real values back into the AI’s reply. From your point of view, it’s mostly invisible; from the AI’s point of view, the sensitive stuff was never there.

I suggest putting it at the top of the readme!

I think it's a very smart system, and could easily extend beyond AI use. Please continue working on it!

Really appreciate your comment!

I put this paragraph in Github project description. Hopefully it's less confusing now.

Great project! I'm looking for something like this but for code: an app that runs locally and obfuscates variables & strings into generic things, so you don't leak code/db structure to LLM companies. Couldn't find anything though so I'm using an LLM to write one.

I’m not sure whether this project is truly practical in its current form. More and more people are using AI agents through CLI tools or desktop applications, so I think you should consider extending it to run locally.

The core is plain JS so it is easy to make a plugin, but I believe there are many plugins already for common coding agents.

I wanted to target the less savvy and non-coder people who treat chatgpt and friends like the new google.

Some clients do that automatically: https://swival.dev/pages/secrets.html

I don't understand this, what's the point?

How does anyone accidentally give personal information to an LLM? Because either you don't in the first place (eh, I guess, most people?) or you do but for some specific, weird reason (I can't think of any but I guess these exist).

It could be useful more as a labor saving device to enable censoring while allowing the human to engage in easy copy-paste. Examples I think of are inclusion of internal DNS names, usernames and human names and email addresses that may be intermixed with log or command output needed to debug some issue.

That said, if these strings are really sensitive and I'm too lazy (or not trusting the human user) to self-censor, I'd not rely on this particular kind of tool for censoring. I'd use local LLMs or cloud LLMs where privacy is part of the contract.

Every time claude inspects `git log` it gets my name and email address.

As @frumiousirc said, it is mainly for the copy-paste case (e.g. you want llm to quickly check out contract terms for you without leaking the parties' identities.)

I have to say I find it amusing how a lot expect people to "not be lazy", people are mostly lazy and even sometimes clueless about how much they leak to BigCorps.

I've found plenty of scenarios where I'm providing logs and it contains a real name, personal info, etc. Stuff like file names, file paths, and little nuggets of information that can lead to an LLM or malicious actor profiling you.

Reply on news.ycombinator.com