Skip to main content

Creating Route

Creating Route

This guide explains how to create routes with different safety configurations.

curl -X POST \
-H "Content-Type: application/json" \
-H "x-javelin-apikey: $JAVELIN_API_KEY" \
-d '{
    "name": "test_route_1",
    "type": "chat",
    "enabled": true,
    "models": [ 
        {
            "name": "gpt-3.5-turbo",
            "provider": "openai",
            "suffix": "/chat/completions"
        }
    ],
    "config": {
        "rate_limit": 3,
        "retries": 3,
        "archive": true,
        "retention": 7,
        "dlp": {
            "enabled": false 
        }
    }
}' \
"https://your-api-domain.com/v1/admin/routes/test_route_1"

Configuration Options​

Basic Settings​

{
    "archive": true,
    "retention": 7,
    "rate_limit": 0,
    "retries": 5
}

DLP (Data Loss Prevention)​

Data Loss Prevention configuration helps protect sensitive information:

"dlp": {
    "enabled": true,
    "strategy": "promptfoo_dlp"
}

Content Filter (Trust & Safety)​

Content filtering allows you to reject requests based on content types:

"content_filter": {
    "enabled": true,
    "error_code": 200,
    "content_types": [
        {
            "operator": "greater_than",
            "restriction": "sexual",
            "probability_threshold": 0.25 // Threshold can be set to 0.25, 0.5, 0.75 or 1
        }
    ],
    "reject_prompt": "Unable to complete request, trust & safety violation detected"
}

Available content restrictions:

  • sexual
  • violence
  • hate_speech
  • profanity
  • crime

Prompt Safety​

Prompt safety helps prevent prompt injection and jailbreak attempts:

"prompt_safety": {
    "enabled": true,
    "error_code": 200,
    "content_types": [
        {
            "operator": "greater_than",
            "restriction": "prompt_injection",
            "probability_threshold": 0.25 // Threshold can be set to 0.25, 0.5, 0.75 or 1
        },
        {
            "operator": "greater_than",
            "restriction": "jailbreak",
            "probability_threshold": 0.25 // Threshold can be set to 0.25, 0.5, 0.75 or 1
        }
    ],
    "reject_prompt": "Rejected from prompt safety policy."
}

Security Filters​

Additional security filters can be enabled:

"security_filters": {
    "enabled": true,
    "content_types": []
}

All safety features support an inspect mode configuration that enables detection of various content types including language patterns, code/markdown syntax, and non-ASCII characters. The SecurityFilters Processor tracks and provides metrics for these security filter matches to help monitor and analyze potential security concerns.