API
One call: a state and any number of typed questions in, calibrated probabilities out. Edit the request and send it. It is answered by the model running in this tab, so nothing you type is sent anywhere.
POST/v1/systemone
valid request · 3 questions · cmd/ctrl + Enter to send
Loading the model into this tab...
Request
- stateobject or text
- Everything the decision depends on. Lead with what matters: very long states are trimmed.
- questionsobject
- Keyed by your own ids. All of them are answered from one reading of the state.
- questions.*.typestring
choice(pick one),score(a level on a rubric) ornoul(is this true?).- questions.*.instructionsstring
- The question, in plain words.
- questions.*.criteriaobject or array
- choice:
key → description. score: level descriptions, low to high. noul: optional{false, true}.
Response
- answers.*.choicestring
- The winning option key.
- answers.*.scorenumber
- Expected level: the probability-weighted average.
- answers.*.noulnumber
- P(true).
- answers.*.probabilitiesobject
- One probability per option, summing to 1.
- answers.*.confidencenumber
- The top probability. Threshold on it to decide what to automate.
- usageobject
- Tokens read. Nothing is ever generated, so
tokens_generatedis always 0.
Errors come back as 400 with { error: { type, message, path } }. More in the docs and cookbooks.