Research
What we measured.
These pages publish the measurements behind Earlyn's design decisions, with the method, the caveats and the data, so you can check the numbers we quote.
- Turkish speech-to-text on a Mac
Turkish speech-to-text on a Mac: Whisper vs Qwen3-ASR.
On 60 Turkish FLEURS clips, Whisper large-v3-turbo made 8.2 errors per 100 words and Qwen3-ASR 1.7B 10.9, on an 8 GB M2 Mac. Method, data and translation notes.Read → - How much disk screen memory takes
How much disk space screen memory takes.
Measured: a busy hour of Earlyn screen memory takes about 9 MB at the default 2,560-pixel capture, and a still screen stores nothing. Resolution, OCR and CPU data.Read → - Local search by meaning
Searching your screen by meaning, on the Mac.
EmbeddingGemma and bge-m3 ranked the right document first for 12 of 12 English and Turkish queries; Apple's NLContextualEmbedding for 4. How Earlyn chose its model.Read →
