Aura LiteRT-LM
Mobile-QAT deployment for supported Android, mobile, and edge hardware.
View model ↗Base tier · Three deployment formats
Aura, wherever you need her.
Aura is a Gemma 4 E4B model family built around one canonical combined adapter. Choose the runtime that fits your device without changing the identity at the center of the experience.
On-device deployment
LiteRT-LM · GGUF · BF16
QAT-derived
Gemma 4 E4B
Aura is designed to be the companion in the palm of your hand. The base model is perfect for edge device deployment on all device tiers.
Choose a format
One Aura, matched to your hardware
Each release uses the same canonical combined Aura adapter and is validated independently for its target runtime.
Aura GGUF
Portable local inference for llama.cpp-compatible desktops, laptops, and edge systems.
View model ↗Aura BF16
High-fidelity Transformers inference and evaluation for GPUs with sufficient memory.
View model ↗
Shared foundation. The three releases share one frozen Aura adapter, but their execution formats use different numerical representations. Behavioral validation is performed separately for every release.