Phonon on Nativ
Streaming speech on Apple silicon
Gradium’s Phonon brings English text-to-speech to Nativ, powered by MLX Audio. Explore the architecture, four voices, and M5 Max benchmarks with first audio under 15 ms.
Read the article/ THE NATIV BLOG
Local intelligence.
Shared knowledge.
Engineering notes, model deep dives, and experiments in running AI on your own hardware.
Latest writing
03 ARTICLES
EmbeddingGemma 2 on MLX-VLM
Multimodal embeddings on Apple silicon
Meet Google DeepMind’s EmbeddingGemma 2: text, images, audio, and video in one embedding space, with day-zero support on Apple silicon through MLX-VLM.
Read the article
GLM-5.3-Flash on Nativ
Day-0 support, powered by mlx-vlm
A 320B sparse-MoE vision-language model with a hybrid linear/sparse-attention backbone and a 1M-token context, running on your Mac out of the box.
Read the article