Metrics Overview

AI Guard exposes an observability pipeline that captures classification activity, redaction actions, agent response times, and user sessions. Metrics flow from the SDK through the AI Guard service to a time-series database or directly to OneTrust AI Governance.

Pipeline Architecture

┌─────────────┐    POST /metric     ┌──────────────┐     Export      ┌─────────────────────┐
│  SDK Client │──────────────────►  │  AI Guard    │───────────────► │  OneTrust AI Gov    │
│  (Python)   │                     │  Service     │                 │  Discovery Platform │
└─────────────┘                     └──────┬───────┘                 └─────────────────────┘
                                           │
                             Auto-generated│
                             classification│
                             counters      │
                                           │
                                    ┌──────▼───────┐     Flush       ┌─────────────────────┐
                                    │  Meters &    │───────────────► │  OTLP Collector     │
                                    │  Aggregation │                 │  (InfluxDB/Grafana) │
                                    └──────────────┘                 └─────────────────────┘

How It Works

SDK submits metrics — Your application calls client.metric() to record agent response times, user sessions, and redaction events
Auto-generated metrics — The AI Guard service automatically counts classification matches per classifier (the ai_guard.classification meter)
Aggregation — The service aggregates metrics internally using OpenTelemetry instrumentation
Export — Aggregated metrics are flushed on a configurable interval to either the OneTrust Discovery Platform API or an OpenTelemetry Collector

What Gets Tracked

Metric	Source	What It Measures
Classification counts	Auto-generated	How often each classifier detects sensitive data
Redaction events	SDK (`client.metric()`)	How often data is redacted or blocked
Agent response time	SDK (`client.metric()`)	LLM agent latency in milliseconds
User sessions	SDK (`client.metric()`)	User interaction and session counts

Data Privacy

📘
Important
The metrics pipeline transmits only aggregated counts and statistics. Prompts, responses, and classified text content are never sent to OneTrust Cloud. The AI Guard service processes all text locally within your infrastructure.

Agent Identity

In OneTrust AI Governance, a unique AI agent is represented by the combination of agent_id + platform. These attributes are required on all metrics and are automatically injected by the SDK client.

Export Modes

AI Guard supports two export modes:

Mode	Destination	Use Case
OneTrust	Discovery Platform API	Production — metrics feed into AI Governance dashboards
OTLP	OpenTelemetry Collector	Development or custom observability stacks (InfluxDB, Grafana, etc.)

See Metrics Exporters for configuration details.

What's Next?

Meter Definitions — Detailed specifications for each meter
Metrics Exporters — Configure OTLP and OneTrust export modes
Observability & Metrics (SDK) — Send metrics from your application