Audex Trace is live on Product Hunt today 🚀
A tiny macOS app that shows what Apple Music is actually playing - Lossless / Hi-Res / Atmos, output sample rate, Playing Next, and skip warnings.
https://t.co/WM9s639Yfw
Audex Trace is live on Product Hunt today 🚀
A tiny macOS app that shows what Apple Music is actually playing - Lossless / Hi-Res / Atmos, output sample rate, Playing Next, and skip warnings.
https://t.co/WM9s639Yfw
Thanks, that’s actually a really good framing.
I agree that concrete failure examples make the value much easier to understand. Output mismatch, Atmos/AAC route differences, and queue skip warnings are probably the clearest demos.
Skip warnings are definitely one of the most “oh, wait, that’s useful” features, but I think the core promise is still: show what Apple Music is actually playing and what your Mac is actually outputting.
First sale for Audex Trace 🎉
It’s just one payment, but it means a lot.
Yesterday it was just a project I was building.
Today it’s a product with a customer.
Thank you to the first person who bought it.
I’ll keep making it better.
New demo for Audex Trace.
The main idea is simple: show what Apple Music is actually playing, what your Mac is outputting, and weird queue behavior before it surprises you.
Product Hunt launch is coming up next.
https://t.co/OBxOvwU9pC
@danielhanchen@UnslothAI nice. only getting ~30 t/s on an m5 max with plain mlx lol. is the amd gap mostly the dynamic quants or the backend? wondering if q4_k_xl beats mlx 4bit on apple silicon
I accidentally started building a C++20 lossless audio codec because I wanted to understand how close a solo implementation could get to FLAC/ALAC.
Now it has LPC, Rice coding, roundtrip tests, malformed input hardening, and my M5 Max is quietly grinding through codec tests.
This is how side projects happen.
https://t.co/65nUdrDSbh
Trace caught two Apple Music queue items that were about to skip.
Then they skipped, and playback jumped straight to the third track.
Queue-level diagnostics are starting to feel real.
944 tokens/sec from a tiny local model on my M5 Max MacBook Pro feels fake.
LFM2.5-350M with MLX 4-bit
Not an API. Just local inference.
Tiny local models are starting to feel instant. @liquidai