Home / Quickstart: your first Jev call in 5 minutes

Quickstart: your first Jev call in 5 minutes

Tutorial·By OpenJev Editorial·Updated 2026-10-03·5 min read

You need an API key from console.typesafe.ai. Then it is one POST. Below is a complete, copy-pasteable example that asks three different types of question in a single call.

1. The raw API call

curl -X POST https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" -d @- <<'EOF'
{
  "state": "Our Stripe integration has been failing
            for 3 days and we can't ship any orders!",
  "model": "jev-latest",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this",
      "criteria": {
        "billing":   "Payment issues",
        "technical": "Bugs / integration",
        "sales":     "Pricing questions"
      }
    },
    "is_urgent": {
      "type": "noul",
      "instructions": "Does this convey urgency?"
    },
    "frustration": {
      "type": "score",
      "instructions": "How frustrated is the customer",
      "criteria": ["Calm", "Frustrated", "Very angry"]
    }
  }
}
EOF

2. The response

Typed answers, ready for code — no parsing, no regex, no "the LLM ignored the JSON schema" fire drill:

{
  "answers": {
    "department": {
      "choice": "technical",
      "probabilities": { "technical": 0.93, "billing": 0.05, "sales": 0.02 },
      "confidence": 0.88
    },
    "is_urgent":  { "noul": 0.999 },
    "frustration":{ "score": 1.62, "confidence": 0.79 }
  },
  "usage": { "input_tokens": 312 }
}

3. The SDKs

  • Python: pip install typesafe-sdk (needs Python ≥ 3.10). Then TypeSafeClient().system_one(state=..., questions={...}) — it reads TYPESAFE_API_KEY from the environment automatically.
  • JavaScript: @typesafe-ai/sdk on npm.
  • Agent skill: Claude Code users can install the official skill: claude plugin install typesafe@typesafe-ai.

4. First-call checklist

Before you trust the output:
  • Keep the state minimal and relevant — irrelevant context degrades accuracy (context rot).
  • Make each question atomic: something an expert could answer in seconds.
  • Write instructions literally — the model does not read between the lines.
  • Read the confidence, not just the answer — it is your safety valve.

Once this works, the next step is wiring it into real logic — that is Patterns, with working code.