How to use Jev

Get a key, then make one narrow call

Create an API key in the TypeSafe console and export TYPESAFE_API_KEY. Install typesafe-sdk on Python 3.10+ or @typesafe-ai/sdk on Node 20+. The first useful call is a question your app already answers badly with rules or a chat completion.

Keep the answer set small and stable. Name the field after the branch your code will take, not after a vague theme.

from typesafe_sdk import Choice, TypeSafeClient

with TypeSafeClient() as client:
    response = client.system_one(
        state={"document": "I was charged twice. Please fix this ASAP."},
        questions={
            "category": Choice(
                instructions="What is this ticket about?",
                criteria={"billing": None, "technical": None, "other": None},
            ),
        },
    )
print(response.choices["category"].choice)

The HTTP contract

If you skip the SDK, the request is still one POST. model selects jev-latest or a pinned version. state can be a string, object, or list of text. questions is a map of named question objects. TypeSafe documents a combined context budget around 64k tokens, with about 32k for state plus the longest question.

• Rate limits published at launch: 250,000 tokens/second and 1,200 requests/minute. Confirm in current docs.
• Batch related questions in one call so they share state cost and return together.
• Put a confidence threshold in your code. Jev makes the branch easier, not risk-free.

curl -X POST https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "Customer: I was charged twice and I am furious.",
    "questions": {
      "topic": {
        "type": "choice",
        "instructions": "What is the issue about?",
        "criteria": { "billing": "money problems", "bug": "broken product" }
      },
      "urgent": {
        "type": "noul",
        "instructions": "Escalate to a human now?"
      }
    }
  }'

What to do with the answer

Branch in application code. If topic is billing and escalate.noul is above your threshold, route to a human. If confidence is low, send the same state to a larger model or a reviewer. Log input, output, version, and the downstream result so you can tell whether Jev is helping the workflow or only looking decisive.