Sqwish Labs provides kettei, a family of fast decision models, alongside OneHot, a platform for serving, prompt-tuning, and fine-tuning structured choices.
The hosted fat and chub models accept text and images; core and dot are smaller text-only releases described as open source. Sqwish reports 93.1% on 231 public JevBench items and a 160 ms UK-service median. Private results remain pending, and public tests carry contamination risk. The support trace and associated workflow-cost illustration are explicitly illustrative, not measured end-to-end savings. OneHot offers prompt tuning and fine-tuning; its training-time claims are the vendor’s.
How you can use it
Try a decision from your own app in the OneHot console, such as choosing a tool or checking an answer against evidence. The hosted models offer a starting point; the smaller text-only releases can be run separately. Keep examples with known outcomes so you can judge whether prompt changes or fine-tuning actually help.
