Guards

Guards are protective rules that inspect requests and responses flowing through the gateway. They can detect sensitive data, block harmful content, and enforce usage limits before requests reach providers or before responses reach clients.

Guard List

The Guards page displays all configured guards as cards. Each card shows the guard’s name, description, type, enforcement level, and whether it is applied to input, output, or both. A toggle switch lets you enable or disable each guard inline.

Creating a Guard

Click Create Guard to open the form:

Field	Description
Name	A descriptive name (e.g. “Block PII in prompts”)
Description	Optional notes about what this guard protects against
Type	The detection or enforcement mechanism (see below)
Level	How the gateway responds when the guard triggers
Applied To	Whether to inspect input, output, or both
Enabled	Toggle to activate the guard

Guard Types

Type	Description
Content Filter	Detects toxic, harmful, or inappropriate content using a configurable toxicity threshold
Rate Limit	Enforces request rate limits beyond the global settings
Token Limit	Caps the maximum token count per request
PII Detection	Scans for personally identifiable information patterns
Prompt Injection	Detects attempts to manipulate model behavior through crafted prompts
Custom	User-defined guard logic

Enforcement Levels

Level	Behavior
Block	Rejects the request or suppresses the response with a 4xx error
Warn	Allows the request to proceed but logs a warning and may trigger alerts
Log	Records the detection in audit logs without affecting the request

Type-Specific Configuration

Content Filter

When the type is set to content-filter, a Toxicity Threshold slider appears. The threshold ranges from 0 (most permissive) to 1 (most strict), with a default of 0.70. Content scoring above the threshold triggers the guard.

PII Detection

When the type is set to pii-detection, a checklist of PII Patterns appears. Select which patterns to detect:

• email — Email addresses
• phone — Phone numbers
• ssn — Social security numbers
• credit-card — Credit card numbers
• address — Physical addresses
• name — Personal names
• ip-address — IP addresses

You can enable any combination of patterns per guard. Create multiple PII guards with different levels if you want to block some patterns and only warn on others.

Applied To: Input vs Output

Target	What is inspected
Input	The user’s prompt and any attached content before it reaches the provider
Output	The provider’s response before it is returned to the client
Both	Both directions are inspected

Applying guards to output is useful for catching provider responses that contain sensitive data or inappropriate content, even when the input was clean.

Editing and Deleting Guards

• Click Edit to modify a guard’s configuration.
• Use the inline toggle switch to enable or disable a guard without opening the editor.
• Click Delete to permanently remove a guard. A confirmation dialog warns that the protection it provides will be removed.

Best Practices

• Start guards in Log mode to understand detection patterns before switching to Block.
• Apply PII detection to both input and output to prevent data leakage in either direction.
• Use content filters on output to catch unexpected provider responses.
• Layer multiple guards — for example, a warn-level PII guard and a block-level prompt injection guard can run simultaneously.