Ask HN: Would you use an AI that remembers everything you've done on your PC? | Hacker News Reader

Ask HN: Would you use an AI that remembers everything you've done on your PC?

2 pointsekalabya4 comments

I've been working on Ghost Widget – an AI memory assistant that captures your screen activity and lets you (or AI agents) search through everything you've ever seen. The problem I'm solving: You spend 20 minutes looking for that article you read last week, or that code snippet you wrote yesterday, or remembering what your manager said in that Zoom call. Your computer has a terrible memory, even though it literally has perfect storage. How it works: Runs in background, captures screen metadata/activity, builds a searchable knowledge base. You can ask "what was that Python library I looked at last Tuesday?" or let Claude/ChatGPT query it directly for context-aware assistance. Early traction: 10 people signed up for demos in the first week with basically no marketing and a landing page I'm embarrassed to show you. Questions for HN:

Privacy: This is obviously the biggest concern. Would local-only storage + open-source code make you comfortable? Or is screen recording a hard no regardless? Use cases: Beyond "finding stuff I've seen," what would make this genuinely indispensable for you? Pricing: Would you pay $20/month for this if it saved you an hour daily? Or does it need to be freemium? Competition: Rewind exists but is Mac-only and just got acquired/pivoted. Microsoft Recall was a privacy nightmare. What would differentiate a new player here?

For context: I'm a 9th grader from India, top 1% on Codeforces, built an AI IDE in 2022 before Cursor was a thing. I ship fast and want to make sure I'm building something people actually want before going deeper.

A very bootstrapped landing page: https://www.ghostwidget.com/

Absolutely not. First of all, any entertainment hours spent on my system are not important. When I am doing something important, it results in some kind of result - a document, an image, research notes.... something. Anything that did not land in such a result means it was not relevant to my end goals. Having an AI store and search information that did not end up being useful would just be a waste of time.

You might have better luck with this idea if you research what types of work deal with actions and data that will always be relevant in the future. I'm not thinking of any, but there probably are such scenarios.

I would not be comfortable with constant screen recording, personally. Local-only storage + open source code wouldn't increase my comfort. My machine would still be compiling a very sensitive database, keeping all that data in one place. That's a bad situation because should my machine be breached, then the whole candy store is robbed. The theoretical benefits don't come anywhere close to enough to make me entertain taking that risk.

But I am not a typical user.

If I want to know... what Python library I looked at, I have browser history; what I did in code, I have Git; who I spoke with, chat apps have searchable history; what commands I ran; I have terminal history;

We have been solving the history problem with computing for decades now. My limited, imperfect human memory is a feature, not a bug. We get very good at the things we do regularly.

App security issues aside, my opinion of a more useful product would be about how can you make the person more efficient, not by doing something for them, but by informing them of how they can grow.

For example: "I noticed you... write many python command line apps; have you tried the python Fire library?"; you are struggling with some python basics; would you like me to schedule some time to practice Python with some of the libraries you use most?"; you do this task regularly; I think you can automate that like this.."

Hard open source with zero network connectivity of any kind (other than to a llama-server hosted on my own network, or maybe an optional paid provider) is the only way I would trust this. Otherwise, this is literally spyware.

Reply on news.ycombinator.com