The URL Analyzer plugin fetches and analyzes content from URLs found in user requests. It detects jailbreak attempts and PII in external content before it reaches your AI application.
Features
- URL extraction: Automatically detects URLs in request content
- Jailbreak detection: Analyzes fetched content for prompt injection attempts
- PII detection: Scans URL content for sensitive information
- Parallel processing: Analyzes multiple URLs concurrently
- Early exit: Stops processing immediately when a threat is detected
- Domain filtering: Whitelist or blacklist specific domains
Configuration
{
"name": "url_guardrail",
"enabled": true,
"stage": "pre_request",
"settings": {
"mode": "protect",
"threshold": 0.7,
"url": {
"timeout": 5000,
"max_content_size": 1048576,
"allowed_domains": [],
"blocked_domains": []
},
"pii": {
"entities": [
"email",
"credit_card",
"iban",
"phone_number"
]
},
"credentials": {
"base_url": "https://api.neuraltrust.ai",
"token": "your-api-token"
}
}
}
Settings
Root Settings
| Parameter | Type | Description | Default |
|---|
mode | string | protect (block) or observe (log only) | protect |
threshold | float | Jailbreak detection threshold (0.0 - 1.0) | 0.7 |
URL Settings
| Parameter | Type | Description | Default |
|---|
url.timeout | integer | Request timeout in milliseconds | 5000 |
url.max_content_size | integer | Maximum response body size in bytes | 1048576 (1MB) |
url.allowed_domains | array | Whitelist of allowed domains (empty = all allowed) | [] |
url.blocked_domains | array | Blacklist of blocked domains | [] |
PII Settings
| Parameter | Type | Description | Default |
|---|
pii.entities | array | List of PII entity types to detect and block | [] |
Credentials
| Parameter | Type | Description |
|---|
credentials.base_url | string | NeuralTrust API base URL |
credentials.token | string | API authentication token |
Modes
Protect Mode
When mode is set to protect, the plugin will:
- Block requests if jailbreak is detected (score >= threshold)
- Block requests if configured PII entities are found
- Return a
403 Forbidden response with details
Observe Mode
When mode is set to observe, the plugin will:
- Log all findings without blocking
- Allow requests to proceed
- Record jailbreak scores and PII detections in observability data
Domain Filtering
Allowed Domains
When allowed_domains is populated, only URLs from those domains will be analyzed. All other domains are skipped.
{
"url": {
"allowed_domains": ["trusted.com", "internal.company.com"]
}
}
Blocked Domains
URLs from blocked domains are skipped and marked as not allowed.
{
"url": {
"blocked_domains": ["malicious.com", "untrusted.org"]
}
}
Blocked domains take precedence over allowed domains. If a domain is in both lists, it will be blocked.
Response Examples
Jailbreak Detected
{
"error": "jailbreak detected in url content (score: 0.92, threshold: 0.70)",
"status": 403
}
PII Detected
{
"error": "PII detected in url content: [email, credit_card]",
"status": 403
}
Use Cases
- Prevent Indirect Prompt Injection: Block malicious content from external URLs
- Data Exposure Prevention: Detect PII in linked documents and pages
- Compliance: Audit external content accessed by AI applications
- Security Monitoring: Track all external URLs referenced in requests
