V4u17 - Your personal, AI-enhanced data storage vault with a low-trust architecture

Hi all, I created this custom GPT/App that I think the community might be interested in called V4u17 (vault). It’s is a personal, encrypted data store for GPTs that enhances the experience of existing GPTs. I’m looking for people to join the private beta, details on the website: https://v4u17.com/. Intro video here: https://www.youtube.com/watch?v=2KsF5uiXtZs . Feel free to cover it on your own video channels if it’s useful, just let me know. Thanks.

I’m not sure if I’m missing something here, can you use this service to fill out forms outside of ChatGPT?

It looks to me that this is storing the data within the app, while encrypted, it all goes plain text when you hand it over to open AI (well through https, but you are still delivering potentially sensitive data (why else would you encrypt it) to a company.) I’m a bit confused what the goal is.

Would you mind explaining?

No, you are not missing anything… you need to trust OpenAI with your data using this app. Currently you can’t use the app outside of ChatGPT (apart from a data export file). Eventually I’d want to make the data portable to other formats, perhaps via an SDK for people to integrate into their own workflows.

The assumption here is that the data you are presenting is/could be useful for the interaction. In the fake pizza store example I use in the video, the pizza store GPT has very basic instructions, it just says that it’s acting as a pizza store and orders require a name and address to deliver it to, then it has a basic menu. When the data is loaded, ChatGPT worked out on its own that that data was appropriate to deliver the order to. I did not program that at all.

So, this sort of turns the current web 2.0 interactions on their head. Currently, UX designers craft the experience that they believe users want - which ends up in our current multi-step checkout experience. In this new world, a business just needs to present the context of the service they are offering and the user decides how to interact for themselves using natural language - every experience is personalised. Also in the video, the return trip to the store is even faster because the previous order could be stored. Nobody programmed that functionality - ChatGPT worked out based on the favorite order data that it could use to complete.

You are correct that you’re offering data to a company, but this data would have been presented anyway in some sort of web page form anyway. At the end of the day you are trusting companies with your data regardless. My GPT/App gives you the control over what data you want to save or present. I use the example of a name and address, but you can save whatever data you want - it doesn’t need to be personal information if you don’t want to share.

Given this is a 1.0 app, it’s fairly basic. But imagine if you can use natural language in the app to be selective about what data you share for an interaction, or better yet, imagine if the GPT service you are consuming could signal to the data store what minimal data is useful for the interaction - then it would only prompt for the data that is missing.

Not sure if any of this resonates - does any of this make sense (I’d like feedback to better land messaging around the value my app offers)? Thanks.

1 Like

Yes, I get that, and it’s a cool idea, but it’s a bit flawed because GPT figured that out, not your App. It looks like you’re selling a solution here that understands your data when it’s really just a storage on your phone, the rest is done by GPT which means the LLM is required to fill out itself.

If you want to create a product that brings the value you are aiming for, the AI part has to be happening within the App itself, which presents you with the problem of a media break: The App can’t interact with the PC i’m ordering my Pizza from. You might have to come up with a two pronged attack, a browser plugin / agent on the device and the App providing the input based on the internally stored data. (Or just skip the App and have people create an account for linking the plugin / any agents to a cloud storage, the data is encrypted after all - if you don’t feel like providing costly storage, have people make use of their own clouds)

I am not expecting businesses to use an LLM, paying for requests, that is capable of figuring out our data (what goes where) on its own, when its much simpler to have the code created for the forms we have been using within seconds.

I get what you are trying to create here though, and the idea is great. I am so wishing to get rid of MFA and passwords, filling out forms etc. But, again IF i understand this correctly, this is more like a POC than a solution?

1 Like

Great idea and prototype! I’ve been thinking about something similar believing we shouldn’t have to deal with products that need our information but a ‘personal agent’ could do this for us instead. It’s a very interesting idea to manage your data in one place and not have many products and services keeping their own collection of knowledge on you as an user.

Reading this again, I believe it would be good for only relevant data to be shared. So you would only permit access to segments of your data. This could be managed by an agent acting on your behalf.

Thanks for the comments @witlist.design and @Matz. I’ve been thinking a bunch about how to improve this service and your feedback is helpful. I did have it on the roadmap about limiting data shared, which seems like a natural thing to do. I’m thinking now that I need to provide the llm interaction for storage client-side, maybe in the app or in a browser extension.

Idea would be that the user would ask to save the data, then all the LLM interaction to save the data and create a schema are hidden from the user (if they want). I could also segment the data by the service you’re interacting with, but also mix in personal data if the user wishes to offer it.

I’m thinking that I could live stream my attempts at improving this service. Much of this work is an experiment on using as much AI as possible to develop. I have a good deal of technical skills, but, for instance, I’ve never programmed a browser extension, so I would rely on ChatGPT, etc. to help me with the basics, if not doing most the coding. I could also expose my product design thinking. I wonder if people would find that interesting - would also give me good feedback in real-time.

2 Likes

I could also maybe try to use multiple AI agents to double check the formatting of the data schema and storage behind the scenes and maybe incorporate AI screenshot analysis to identify fields that could be filled in to see if any stored data matches it. Some good examples of people using visual tagging to achieve this.

1 Like