QuickQuill updates: Two new features, both about search.
Semantic search
You can now search your meeting history by meaning instead of exact words. It all runs on-device, and it works across languages, so a Japanese query can find an English meeting.
Some notes on how this got built, because it didn't go the way I planned. My first version used Apple's built-in embedding API (NLContextualEmbedding), mainly because it needs no model download. I really wanted that to work. But it's a general-purpose model, not tuned for retrieval: short filler lines like "yeah, okay" matched almost any query, and the similarity scores moved around so much that I couldn't find a threshold separating real hits from noise. I tried the usual tricks (centering the vector space, weighting by length) and they helped, but on my real transcripts it was still a coin flip. So I gave up on it and switched to Qwen3-Embedding-0.6B, a model actually trained for retrieval, running locally through llama.cpp. That costs users a one-time 640 MB download, which I was reluctant about, but the results on my own messy meeting data were good enough that I stopped second-guessing it.
Global search launcher
Press Option+Space in any app and a Spotlight-style search bar pops up. Pick a result and QuickQuill jumps to that moment in the transcript. You can change the shortcut, and Esc puts focus back where you were.
To update, Settings > General > Software update. Semantic search is off by default. Turn it on in Settings > General and download the model from there.
Replies
Dial
the write-up about ditching NLContextualEmbedding is the useful part here, more people should share the "the free option didn't work" story instead of just shipping the win. question on the opt-in default though - since Qwen3-Embedding-0.6B is apparently working well on your own messy data, is turning it off by default purely about the 640MB download, or is there an ongoing CPU/battery cost during indexing that makes it something people should deliberately opt into rather than just a one-time cost? and when someone does turn it on, does it index the whole existing meeting history retroactively or only meetings recorded from that point forward?
QuickQuill
@galdayan
Thanks! It's mainly the 640 MB download. I didn't want a background download that big to be the default. Indexing runs once in the background after you turn it on, and after that only new recordings get embedded, so there's no ongoing cost you'd notice.
And yes, it's retroactive. Your whole existing history gets indexed, so old meetings become searchable by meaning too.
Dial
@taisei_ide good, that's the answer I was hoping for - one-time cost with a real retroactive payoff makes the opt-in trade pretty easy to say yes to. thanks for the detail