Notes for November 17-22

It’s now cold enough for me to enjoy being inside, but sometimes I would prefer to have a bit more variety and not be stuck in front of a screen all the time. Still, there’s been some progress on various fronts this week…

Someone famously wrote, at the dawn of this new age that is all of three years behind us now, that AI would either be a way to get more things done or a massive multiplier of unfinished projects, and right now it looks like the latter.

Over the past week or so, I’ve built:

• A working, but still woefully incomplete PyObjC applet to replace the bloated and useless Elgato StreamDeck App.
• A roughly 75% complete port of this site’s static generator to TypeScript running on bun.
• An audio transcription helper that re-processes recorded calls (or takes existing transcripts) and that helps me create meeting notes with the level of detail I need (I find the default Teams transcripts to be overly generic).
• A FastAPI conversion of the TRMNL self-hosted server, because I would love to not run a full Chromium browser to render a few measly bitmaps.
• An autogen agent editor for quick prototyping.

None of them are what I would call finished, but they are certainly useful–although I am still on the fence regarding whether AI has actually saved me any time…

After years (in fact, pretty much around a decade) of using Feedly as a back-end for Reeder (the “Classic” version, not the low-attention-span, overly “social” modern one that I am really not a fan of), I decided to set up FreshRSS as a personal news aggregator–not because I like its web UI (I don’t, and thank goodness I don’t have to use it all the time), nor because I like its implementation (it’s written in PHP, of all things), but because it works well enough as a Reeder back-end and affords me other advantages:

• It provides me with a single back-end to gather private RSS feeds and have them easily available on all my devices.
• It lets me specify per-feed proxies (which is handy when coupled with tor for geo-locked content) and readability-like full text extraction from stingy content providers.
• I can finally use some Android and Linux apps that don’t support Feedly (like Newsflash, which is what I use on Fedora).
• I was finally forced to remove dozens of dead feeds off my blogroll. It was somewhat depressing to see that pretty much nobody from the smartphone boom era still has a blog, but
• It’s one less thing hanging off my Google account.

I have also published the source code for my feed summarizer, which also provides some pre-processed feeds to FreshRSS. With both inside the same machine, I can now have all my private daily bulletins and conventional feeds aggregated in one place, served up to my iPad, Nomad, etc., and my read status synced all my devices.

I’ve been thinking about improving the summarizer with vector search to have better semantic grouping (right now I’m just coaxing gpt-5-mini to do it), but I’ve come to the conclusion that generating embeddings is just too much to ask for the tiny machine running the show right now, and that it doesn’t warrant the additional complexity. At least for now.

As a minor note, I upgraded all my Proxmox nodes to 9.1 without any issues. I haven’t yet tested the new OCI support, but while spelunking through the forums I found out about ProxMobo, which is a refreshingly subscription-free and well-designed iOS app for managing Proxmox clusters, including support for receiving iOS notifications tied to Proxmox’s native alerting system.

I am still struggling with printing some functional parts, so I again spent entirely too much time not just reassembling and recalibrating both my printers but also planning some upgrades in order to make do.

That is getting old really fast, so I’m definitely considering getting a new printer.