When you do a search or add an interest, I do call their API to embed the query. However, for interests you've configured, I store the embedding so I don't need to call their API each time

Glad to hear it! I’m using their API to generate the embeddings but I’m storing them myself in SQLite

In both the RSS and Atom specs there is a source property on the item that I think is meant for this purpose

(And meant to mention that Scour also turns every user feed and interest-specific feed into an outbound RSS feed!)

Better Air Purifiers (www.jefftk.com/p/better-air...)

Super interesting -- and creative DIY solution!

Using 'Slop Forensics' to Determine Model Ancestry (www.dbreunig.com/2025/05/30/u...)

Such a neat technique!

RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search (dl.acm.org/doi/pdf/10.1...)

I ❤️ reading about other binary embedding quantization techniques!

I came across the paper Harnessing the Universal Geometry of Embeddings (arxiv.org/abs/2505.12540) and The Platonic Representation Hypothesis (arxiv.org/abs/2405.07987). It's pretty wild that different models converge on similar representations of the same data!

Thanks for letting me know! I'll DM you to get more details and figure out what's going wrong

Thanks! It was quite a bit more work than I anticipated 😅

Thanks @8hantanu.net for the inspiration to think of it as a stack! 😊

I’ve heard good things about Refactoring UI from the Tailwind folks.

Haven’t gotten around to reading it yet so I’ve been using picocss.com 😅

Glad you liked it!

I've currently got it running on a shared 8-CPU, 2GB RAM Fly.io instance. That'll definitely change over time but it's been good enough so far.

Thanks Ade!

I briefly looked at both of those but a) they didn’t seem that mature so I was worried about breakage and b) wanted to avoid shipping a big wasm bundle. I imagine the reactivity might feel nicer than the htmx model though