Exploring Portable Context for AI Tools

Hi Nicolae, I’m here to help someone explore an idea they’re working on, and I’ll ask questions to dig deeper. Let’s start simple: what’s the core idea that’s been on their mind lately?

I think you mean what the core idea that's been on my mind.

Exactly, what’s the core idea that’s been on your mind?

So, I am thinking a little bit about just about context in general and interoperable context, I guess, you know, like, it's, I'm thinking about ways to move context between various LLM tools and how to make that context layer portable. But even, like, you know, even showing up to an app like this one, it's a bit intimidating to start talking with you about things from scratch when you know very little about me. And so, I'm thinking about, like, what would almost like a sign-in with context look like where I could be carrying access to those various things or I could even app mention and give you access to specific things, almost like a sign-in context layer.

That’s really interesting. So it sounds like you’re exploring how to move context seamlessly between tools and even bring it into conversations. What part of that feels most challenging or exciting to you right now?

I mean, I think there's like some things that would be relatively easy to give access to. So, for example, I could conceive of giving access to like my banking data, my email, my Notion, so and I could even imagine having like a bunch of derived data that you could easily get access to on text. I could even imagine giving access to some of my browsing history in some organized ways. So we could like talk about articles or tweets I've looked at. Um, but I think that, uh, yeah, then there's other, uh, context that's a little bit. More tricky, like, I think all the conversations that I've had with Claude and chat are interesting and. I'd be curious if there's ways using the. Api's with the files on the file system for these different service providers. So, like, almost have a portable. Generated context, or even to be able to. Move that context across the tools. I. 11 product, I even conceived of at 1 time was something around, uh. Migrating your memory from. From to open, which seems doable. So, yeah.

It sounds like you’re thinking about making those personal AI memories portable. What do you think would be the biggest hurdle in moving that kind of deep conversational history between tools?

I mean, I think it's just like sensitive data. So I think there's like, I think security is like a huge thing on this and I think ease of accessing that data is another aspect. I don't think that Couch API gives you an API to access the computation history for a user. And I'm not sure about Entropiq and Cloud. And then I think it's also just like a lot of unstructured data. So there is also some question of like, if I was to sign in with something, how would that be exposed and how would you make that easy to use?

That makes sense. So how might you imagine structuring that unstructured data to make it easier to access and share across tools?

Well, interestingly, right, it's like, hey, I want to bring the context from all my other tools to this tool, maybe the context and the workflows from all my other tools to this tool. And this tool also has some context and some workflows. And so it's like, how did these things get stitched together, right? Like in this case, the context and workflow here had something to do with interviewing me. So in order to do a good job interviewing me, you might want to know a lot more about me than you really know right now. And you could get some of that from the web. So there's also a ton of public web context as well. Anyways, yeah. And that's, I think, yeah, around, there's probably this idea of signing up and gathering a bunch of context from the web about someone, about a customer, for example. And that's its own little kernel, too, that you could start with just that as an easy starting point. Yeah. Yeah. Yeah. Yeah. Yeah.

I see. So are you thinking that starting with public web context—like what’s already out there about someone—might be an easier first step before bringing in private or sensitive data from other tools?

Yeah, I might start with something like that, which is very reminiscent of Clay for me. And it's a product we really try to build a ton. And I might start with that and build something like that for another one of the tools that I'm using to see how that feels and try to build it as a standalone product as well. Almost like a contact layer for signups. Yeah, and we could do that via an API to start with, I guess.

So if you were to test that out, which tool would you pick first to integrate this context layer with, and why?

I could imagine doing it on top of this listening product, but I'm also building a finance application for small businesses, and I think it would be really interesting there.

That’s intriguing. Between the listening product and the finance application, which one do you think would show the value of portable context faster?

I think probably the finance application if I'm if I'm honest with myself And so yeah, that'll be I'm gonna slate that on the agenda for doing a little prototype this week

That sounds great. What’s one key piece of context you’d want to capture first in that prototype, to really make it shine?