Jev (TypeSafe AI)

Make fast, structured decisions with TypeSafe AI Jev

Overview

Jev is TypeSafe AI’s System One model. Instead of generating text, it evaluates a state against typed questions and returns calibrated probabilities, so a following step can branch on a number rather than parse a sentence. It fits classification, routing, rubric scoring, and guardrail checks.

Key Features

• TYPESAFE_JEV_EVALUATE
  • Evaluate one state against many questions in a single request
  • Three question types:
    • noul — yes/no. Returns noul, the probability (0–1) that the answer is yes
    • choice — pick one option. criteria maps each option to a description (or null). Returns choice, probabilities, confidence
    • score — rate along ordered levels. criteria is an array of at least two level descriptions. Returns score (can land between levels), legend, probabilities, confidence
  • state accepts plain text or a JSON object/array
  • model defaults to jev-latest. Pin a version such as jev-1.13.0 if you tune confidence thresholds
  • Rate-limit (429) and overload (529) responses are retried automatically

Authentication

This tool requires your own TypeSafe AI API key; Jinba API credit is not available for it. See the TypeSafe AI quick start to obtain one.

Note: Treat API keys as sensitive information and never commit them to public repositories.

Limits

• Text only, with a budget of 64k tokens per request (32k for state plus the longest question)
• English is the most accurate language. For other languages, check confidence before acting on an answer

Example: Triage a Support Ticket

- id: triage
  tool: TYPESAFE_JEV_EVALUATE
  input:
    - name: state
      value: "Help! My payouts have been failing for 3 days."
    - name: questions
      value: |
        {
          "is_urgent": {
            "type": "noul",
            "instructions": "Does this convey urgency?"
          },
          "department": {
            "type": "choice",
            "instructions": "Which team should handle this?",
            "criteria": {
              "billing": "Payments, invoicing, refunds",
              "technical": "Bugs, outages, integrations",
              "sales": "Pricing, upgrades, new accounts"
            }
          },
          "frustration": {
            "type": "score",
            "instructions": "How frustrated is the customer?",
            "criteria": ["Calm", "Frustrated", "Very angry"]
          }
        }
  config:
    - name: api_key
      value: "{{secrets.TYPESAFE_API_KEY}}"

Answers come back under the same ids, e.g. {{steps.triage.result.answers.department.choice}} or {{steps.triage.result.answers.is_urgent.noul}}.