Frontier AI. Compressed 10×. On the device you already own.

Wisteria. Frontier capability on consumer hardware.

Our 1.58-bit ternary language model series, recovered from frontier FP16 parents via continual pretraining and distillation. Wisteria-122B-A10B is the current proof — a 122-billion parameter MoE in 24 GB, running offline on a MacBook. The series scales from 125B to 1T parameters.

"Health data is the most private data people own. We're proving that using state-of-the-art AI doesn't require sending it to the cloud."

Ovara. Cycle prediction without the cloud.

Your most private health data stays exactly where it should — on your phone. On-device cycle prediction with zero cloud dependency and differential privacy. Currently in active research with early models training on synthetic data.

Hush. Baby tracking with sleep prediction.

Logs feeds, nappies, and milestones — then predicts sleep regressions before they happen. Built by a pharmacist who understands developmental science.

Sana. Chronic condition flare prediction.

Tracks mood, energy, symptoms, and lifestyle — then warns you before a flare hits. For migraines, IBS, eczema, and joint pain.

Research.

News.