Benchmarks & research
Tachara AI Lab's Decisio turns unmodified open models, by default Google's Gemma 4 31B, into a self-hosted decision server that gives a probability for every option.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Each option gets a letter, and the answer is the model's probability for each letter at one position, so nothing is generated or fine-tuned. It speaks TypeSafe's published /v1/systemone format but is independent of TypeSafe. The default base needs a 96 GB graphics card; a smaller Gemma build is offered through Ollama for 16 GB laptops.
Decisio is for teams that want Jev-style answers while keeping their text on their own machines. You could ask three things about every support ticket at once: which team should take it, how urgent it is and how badly it affects the customer. It reads the ticket once, so the extra questions are fast. It can also learn one of your regular questions from examples you have already sorted.