Glossary

Catalog

Consumer

An internal system or application that makes requests to one or more Resources.

For more details, see Consumers and Owners.

Owner

An organizational entity (team, department, or individual) responsible for one or more Consumers.

For more details, see Consumers and Owners.

Provider

An external or internal organization that offers one or more AI Resources.

For more details, see Providers and Resources.

Resource

An external or internal AI service endpoint that responds to requests from one or more Consumers.

For more details, see Providers and Resources.

Budgeting

Budget

A policy that governs the usage of one or more Resources by Consumers.

For more details, see Budgets and Budget Templates.

Budget Template

A predefined configuration pattern for creating standardized Budgets.

For more details, see Budgets and Budget Templates.

Guardrails

Guardrails

A configurable control mechanism that evaluates AI interactions against defined risk criteria to detect and prevent potential security, privacy, or operational risks.

For more details, see Managing Guardrails.

Detection Rules

Specific patterns or conditions that, when matched in AI request or response content, trigger a guardrail event. Rules can be based on pattern matching (regex), ML classifiers, or other detection methods.

For more details, see Creating Detection Rules.

Events

Occurrences when a guardrail's detection rules match content in an AI interaction, indicating a potential risk that requires attention or action.

Enforcement Modes

Monitoring Mode

A guardrail operation mode that logs violations without blocking requests, used for observation and fine-tuning guardrails.

Enforce Mode

A guardrail operation mode that actively blocks requests that violate the guardrail rules based on configured severity thresholds.

Default Guardrails

Pre-configured guardrails provided out-of-the-box that address common AI risks based on industry standards like FINOS and OWASP.

Custom Guardrails

Organization-specific guardrails created to address unique security, compliance, or operational requirements not covered by default guardrails.

False Positive

An incorrectly identified guardrail event where legitimate content is flagged as violating a rule.