AI Command Center
The AI Command Center is the operational hub for AI Governance — a live catalog of every AI agent discovered in your project, with health, cost, latency, and risk at a glance. It's where you go to answer "what AI do we have running, and is any of it misbehaving?"
Enterprise feature. Requires the Owner or AI Steward role. Open it from AI Governance → AI Command Center.
How agents are discovered
Agents are derived automatically from your trace data — each distinct AI workflow reporting traces appears as an agent. There's nothing to register manually: instrument your app (see Integrations) and agents populate the catalog as traffic arrives.
The agent catalog
The main view is a catalog of agents. Each agent card/row shows:
- Status — derived from the agent's risk score (see below)
- Requests, cost, and latency for the selected time range
- Model(s) the agent uses
- Integration badges — linked Slack channel, Jira project, Linear team, ServiceNow, or PagerDuty service
- Risk score — a 0–100 composite signal
Status badges
Status is derived from the agent's risk score so you can triage at a glance:
| Status | Meaning |
|---|---|
| Healthy | Low risk signal |
| Warning | Moderate risk — worth a look |
| Critical | High risk — needs attention |
| Inactive | No activity in the selected window |
Risk score
The risk score (0–100) is a composite of detection counts (PII / PCI / PHI / ethics / safety), compliance violations, drift signals, and error rate. It powers both the status badge and the Risk Dashboard.
Filtering and search
Narrow the catalog to what matters:
- Status — Healthy / Warning / Critical / Inactive
- Cost range — buckets computed from your actual spend distribution (they recalculate per time range)
- Time range — last 1 / 2 / 3 / 7 days
- Model, tags, and integration
- Search by agent name or model
- Sort by last active, requests, cost, latency, or risk score
- Switch between grid (visual cards) and list (compact table) views
Agent detail panel
Select an agent to open its detail panel:
- 24-hour metrics — hourly time series for requests, cost, latency, and errors
- Cost projection — actual spend plus a forecast scaled to a high-volume baseline, so you can see what the agent would cost at scale
- Compliance violations — unapproved-model usage (24-hour and 7-day counts) with the last-violation timestamp
- Detection breakdown — PII / PCI / PHI / ethics / safety detections attributed to the agent
- Integrations — the Slack/Jira/Linear/ServiceNow/PagerDuty destinations mapped to this agent
Unapproved-model violations come from the AI Inventory: if an agent uses a model that isn't approved for your organization, it's flagged here.
Typical workflow
Scan for problems
Open the Command Center, set the time range, and sort by risk score (or filter to Critical) to surface the agents that need attention.
Drill into an agent
Click an agent to open the detail panel. Review its 24-hour metrics, compliance violations, and detection breakdown.
Act
Follow the signal: review detections in AI Violations, approve or block models in the AI Inventory, or open the Risk Dashboard to assess and remediate drift.
Next steps
- Risk Dashboard — assess risk across four dimensions and triage drift.
- AI Inventory — approve the models agents are allowed to use.
- AI Violations — review what your detections caught.