State-machine decisions — branch your workflow on typed answers

Workflow engines choke on LLM output because it is prose; they thrive on decisions because they are values. A decision model call is the missing edge in your state graph: state in, typed question, branch on the answer.

Why decision models

• Pure-JSON request / typed-answer response
• No output parsing, no format retries
• Deterministic branching for pipelines
• Per-request model choice across the catalog

The request is pure JSON (model, state, questions) and the response is a typed answer block — no schema-wrangling, no retries-on-bad-format. Latency stays in the milliseconds-per-decision range.

Runnable example

curl https://api.easyjev.online/v1/systemone \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "typesafe/jev-latest",
    "state": {"user": {"plan": "pro"}},
    "questions": {
      "is_premium": {"type": "noul", "criteria": {"true": "…", "false": "…"}}
    }
  }'