typesafe-ai
TypeSafe AI
Jev is a decision model, not a text generator. One request carries a state (string, JSON object, or array of text) and a map of typed questions; every question is answered in parallel against the same state in roughly 100 ms, and every answer comes back as a calibrated probability distribution. It never writes prose, code, or explanations, accepts text only, and is not fine-tunable. Code owns control flow: rules, math, dates, counting, and side effects stay in code, and Jev supplies narrow judgments over unstructured text.
Request shape
POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer $TYPESAFE_API_KEY
{ "state": ..., "model": "jev-latest", "questions": { "<id>": {...}, ... } }
| Primitive | Asks | criteria |
Answer fields |
|---|---|---|---|
choice |
Pick one of N unordered options | { option: description | null }, up to 255 |
choice, probabilities, confidence |
score |
Position on an ordered rubric | array of 2 to 10 level descriptions, low to high | score (probability-weighted mean of level indices, fractional), legend, probabilities, confidence |
noul |
Probability a yes/no statement is true | optional { true: ..., false: ... } |
noul in [0, 1], no confidence field |
Question ids are for your code and are never sent to the model. instructions and every criteria value accept a string, JSON object, array, or null, so structured descriptions (what, not_for, examples, any names you choose) are allowed anywhere. Full schemas, errors, limits, and model ids: references/api.md.