Restore original 4-language G2P (document-level regression in the v7 G2P; 85 vs 15 audible edits on the 145-line document suite)

Fully quantized Core AI bundle (macOS only: iPhone GPU compilation runs out of memory on the int8 vector estimator)

Default Core AI bundle: int8 frontends + fp32 vector/vocoder (full-int8 OOMs iPhone GPU compilation; moved to coreai-int8)

Keep the fp32 Core AI bundle available as coreai-fp32

Make the int8-quantized Core AI bundle the default (g2p/duration/context/vector int8, vocoder fp32; 384-frame vocoder bucket excluded pending compiler fix)

Add Scylla's Band Core AI bundle for iOS and macOS 27 (384-frame vocoder bucket excluded pending re-export)

Scylla's Band v1: G2P v7 (7 languages), b64/b128 latent buckets, model_version 1

Add optional INT8 ONNX bundle

Initial public release