Personal Privacy Settings

Control how LangMart protects your sensitive information when using AI models.

Getting Started

Access your privacy settings at Settings → Privacy & Security.

Content Protection Toggle

The main toggle enables or disables all PII detection for your account.

State Behavior
Enabled (default) All sensitive data is automatically detected and redacted
Disabled No protection - sensitive data may be exposed to AI models

[!WARNING] Disabling content protection is a security risk. Your API keys, passwords, and other sensitive data will be sent directly to AI providers without redaction.

Per-Type Configuration

When content protection is enabled, you can configure detection for each data type individually:

Type Description
API Keys OpenAI, Anthropic, Google, Groq, DeepSeek API keys
Credit Cards Payment card numbers
SSN Social Security Numbers

Medium Severity

Type Description
Email Addresses Personal and work email addresses
Phone Numbers Phone numbers in various formats

Managing Settings

To Enable/Disable a Type

  1. Go to Settings → Privacy & Security
  2. Scroll to "Configure Detection by Type"
  3. Toggle individual types on/off
  4. Click Save Settings

To Reset to Defaults

  1. Go to Settings → Privacy & Security
  2. Click Reset to Defaults
  3. Confirm the reset

All types will be re-enabled with default settings.

How Detection Works

Your Message: "Use my API key sk-proj-abc123xyz..."

          ↓ PII Detection

Safe Message: "Use my API key sk-proj-7f8a2b1c..."

          ↓ Sent to AI

AI Response: "I'll use the key sk-proj-7f8a2b1c..."

          ↓ Restoration

You See: "I'll use the key sk-proj-abc123xyz..."

The original key is never sent to the AI provider. A random placeholder is used instead.

Viewing Detection Events

After guardrails detect sensitive content:

  • Results appear in your Request History (/requests)
  • Organization admins can view all events in Guardrail Events

Organization Overrides

[!NOTE] If your organization has PII enforcement policies, some settings may be locked. Enforced types are applied silently and cannot be disabled by individual users.

See Organization Enforcement for details.