Thinking about personal OS. More like a personal dashboard but it could expand from there.

I set up a local node API authed into Gmail and GCal. Showing unread and today's calendar. Working on building out some basic timeblock capabilities.

One move I like is adding a top banner for the current activity - that goes back to ideas we talked about at Daylight.

Using GCal (by way of AI-generated API code) is interesting. Big benefit is I have the escape hatch to go do things in the regular interface if needed, and similarly access on things like iPhone apps. Another side benefit may be that it's worked out the complications of events - like a recurring event where you want to make a modification to a one-off. Though I'll likely still have to deal with some of the complications involved in that as I build out ways to interact with it.

One thing I'm hopeful about with email is the ability to reference an email from other parts of the system - particularly todos. Emails often hang around as rough todos. But I think I could build something linked to the ID that would then render the emails inline.

I'm thinking of Alexander Obernaur's talking about itemized OSes. And Szymon's timeblocking/timetracking projects.

And Obsidian's file over app approach - although there's a knotty piece there. I want a database to make things more flexible (could possibly layer a database over the filesystem I suppose). But I want one with a lot of inspiration from files - I think lots of simple entries with text, createdAt and type - and then I'll need to layer query helpers (like embeddings of the text on top). The type part is challenging. I'm thinking of things like recipes where I don't necessarily want to be throwing them into the vector results all the time... Type should do it for that. But then there's an architecture question of how to flexibly have extra fields...

Actually recipes seems really pretty doable, possibly there is more of a dilemma across notes, todos, emails and events.

Notes are probably thinking I want to revisit - great candidates for embedding search while I'm writing other notes - associative. Todos are probably ephemeral. I think the question is - do todos belong to notes? Maybe that is where Obsidian and the other approaches have it figured out. A daily note every day - notes can either be day based or project based...

And then one about events? Notes could be attached to events. Could todos be attached to events? Maybe the main question is do todos live independently? Lesson of Obsidian and LLM stuff is maybe that todos written as plain text are powerful and then you roll up from there.

Another big piece is media - models can understand media but it's expensive - I'm thinking I'll build out a media library interface - uploading to AWS using the key as source of truth (or should I look into hashing the content?). When they get uploaded they also get fed into an AI (probably gemini) to be described. And we embed that description as well. Those then live in postgres with the vectors...

Then a blog post with media in it could be pieced together on the fly (that is a reason to use the AWS key since it will be the URL anyway). With the text inserted - then embeddings of that blog post would be strong as well as an agent reading it.

One thing I am thinking is that I don't want an agent to always be assembling 'briefings' out of this stuff. I want to see the stuff - emails, events. I just want better, less disruptive interfaces to use them in.