Overview

The Jev connector enables your AI Colleagues to integrate with TypeSafe's Jev model, facilitating structured decision-making such as yes/no checks, classification, routing and scoring within automated workflows.

Jev is TypeSafe's evaluation model, designed to assess content against typed questions and return structured answers with probabilities and confidence scores, rather than free-form text. The Jev connector allows Leena AI to make fast, consistent judgments — for example, whether a request is urgent, which team should handle it, or how frustrated an employee is — that AI Colleagues can act on directly.

API Details

Leena AI integrates with Jev via REST APIs.

Documentation link: https://docs.typesafe.ai/api

Setup

The Jev connector uses API Key authentication.

Prerequisites

Before setting up the Jev connector, ensure you have:

• A TypeSafe account (or use Leena AI's built-in authentication)
• Access to the TypeSafe Console for API key generation (optional)
• Understanding of the question types supported by Jev (Noul, Choice and Score)
• Access to your Leena AI workspace with connector management permissions

Get credentials

Here is how to create an API key in the TypeSafe Console:

1. Sign in to TypeSafe: Log in to your account at TypeSafe Console
2. Navigate to API Keys section: Open the API keys page in the console
3. Create New API Key:
   • Create a new API key
   • Name your key: Use a descriptive name like "Leena AI Integration" or "Production Backend"
4. Save and Note Credentials:
   • Copy your API key immediately
   • Store the key securely

Add connection

Here is how to add a connection on Leena AI:

1. Log in to your Leena AI workspace
2. Navigate to Settings > Integrations
3. Search for "Jev" and select it from the list to add its new connector
4. Start configuring the connector:

Model: Choose from available Jev models:

• jev-latest: Always uses TypeSafe's current production Jev model (default, recommended)
• jev-preview: TypeSafe's preview model, for testing upcoming changes before they reach jev-latest
• jev-1.13.0: A fixed model version, for results that stay consistent over time

Auth type: Choose between:

• API key: Use your own TypeSafe API key
• Leena AI auth key: Use Leena AI's managed authentication (no additional configuration required)

API key: If using your own key, paste your TypeSafe API key in the configuration field

5. Test Connection: Click Test connection to verify that Leena AI can reach TypeSafe with the selected authentication
6. Save Configurations:
   • Ensure the key is valid and active
   • Click Save to complete the connector setup
   • The connector will be saved and ready to use

Actions

The following actions are supported for the Jev connector:

Evaluate

Evaluates content (the state) against one or more typed questions and returns a structured answer for each question. The Agent can leverage the tool (workflow), which has been designed to classify, route, score or check content — such as employee messages, tickets or form data — before deciding what to do next.

Input Parameters

Here are the input parameters required to set up this action:

Mandatory

Question Structure:

Each question in the questions array must follow this structure:

Question Types:

Confidence Guidelines:

These ranges are a starting point. Adjust them to suit how critical the decision is.

Here is a sample JSON input:

//Yes/No Check (Noul)
{
  "payload": {
    "state": "My salary hasn't been credited for 3 days and I can't pay rent. Please help ASAP.",
    "questions": [
      {
        "id": "is_urgent",
        "type": "noul",
        "instructions": "Does this message express urgency?",
        "criteria": "{\"true\": \"Urgent\", \"false\": \"Not urgent\"}"
      }
    ]
  }
}

//Routing (Choice)
{
  "payload": {
    "state": "I can't log in to my laptop after the latest update.",
    "questions": [
      {
        "id": "department",
        "type": "choice",
        "instructions": "Which team should handle this request?",
        "criteria": "{\"payroll\": \"Salary, payslips and deductions\", \"it\": \"Devices, access and software\", \"hr\": \"Leave, policies and benefits\"}"
      }
    ]
  }
}

//Sentiment Rating (Score)
{
  "payload": {
    "state": "This is the third time I'm asking about my reimbursement. Nobody has replied.",
    "questions": [
      {
        "id": "frustration",
        "type": "score",
        "instructions": "How frustrated is the employee?",
        "criteria": "[\"Calm\", \"Frustrated\", \"Very angry\"]"
      }
    ]
  }
}

//Structured State with Multiple Questions
{
  "payload": {
    "state": "{\"ticket_id\": \"TKT-10482\", \"subject\": \"Salary not credited\", \"message\": \"My salary hasn't been credited for 3 days and I can't pay rent. Please help ASAP.\"}",
    "questions": [
      {
        "id": "is_urgent",
        "type": "noul",
        "instructions": "Does this ticket express urgency?"
      },
      {
        "id": "department",
        "type": "choice",
        "instructions": "Which team should handle this ticket?",
        "criteria": "{\"payroll\": \"Salary, payslips and deductions\", \"it\": \"Devices, access and software\", \"hr\": \"Leave, policies and benefits\"}"
      },
      {
        "id": "frustration",
        "type": "score",
        "instructions": "How frustrated is the employee?",
        "criteria": "[\"Calm\", \"Frustrated\", \"Very angry\"]"
      }
    ]
  }
}

Response

Upon successful evaluation, the action returns:

• An answer for each question, returned under its Question ID
• Question type for each answer (noul, choice or score)
• For Noul: probability that the answer is yes (0 to 1)
• For Choice: the selected option, the probability of each option and a confidence score (0 to 1)
• For Score: the probability-weighted score (which can fall between levels, e.g. 1.3), a legend mapping each level index to its label, the probability of each level and a confidence score (0 to 1)
• Model version used for evaluation
• Token usage information (input tokens and output tokens)

Here is a sample JSON response:

{
  "model": "jev-1.13.0",
  "answers": {
    "is_urgent": {
      "type": "noul",
      "noul": 0.96
    },
    "department": {
      "type": "choice",
      "choice": "payroll",
      "probabilities": { "payroll": 0.94, "it": 0.01, "hr": 0.05 },
      "confidence": 0.9
    },
    "frustration": {
      "type": "score",
      "score": 1.3,
      "legend": { "0": "Calm", "1": "Frustrated", "2": "Very angry" },
      "probabilities": { "0": 0.02, "1": 0.66, "2": 0.32 },
      "confidence": 0.71
    }
  },
  "usage": { "input_tokens": 412, "output_tokens": 38 }
}

Error Handling

Each request times out after 60 seconds.