strands-decider is a 2B decision model. It doesn't write text: it picks between options, answers yes or no, and rates on a scale, and every answer comes with a calibrated confidence. Write a situation, ask it questions, and see how sure it is.
The marker on each gauge is the model's confidence. Past the line at 0.9, its answers are right about 95% of the time on tasks it hasn't seen; below it, check with a person or a bigger model.
Ask the model to see its answers here. The first request after a quiet spell starts a GPU and can take about a minute.
One POST /v1/systemone takes a state and any number of named questions. The request on the right is the one the form above sends; edit the form and it updates.
noul answers yes or no and returns P(yes).choice picks one named option and returns each option's probability.score rates on an ordered scale, low to high, and returns the expected level.The wire format is the same as strands-decider serve, so existing clients work by changing the URL. Without an API key, requests are limited to a few thousand characters.