Skip to main content

Agents

The Agents section gives you visibility into what your organization's AI agent team is doing at the policy boundary. It surfaces two feeds — policy decisions and agent actions — scoped to your organization.

note

This view is references and metadata only by design: it shows content references and metadata (such as a content reference or a target reference), and never stores or displays raw agent or prompt content.

Accessing Agents

Open Agents from the dashboard sidebar. Viewing agent activity requires the Audit Viewer permission.

The Two Feeds

The page has two tabs:

Policy decisions

Each row is a verdict recorded when an agent's traffic was evaluated at the LLM/policy boundary. Every decision shows:

ColumnDescription
TimeWhen the decision was recorded
AgentThe agent slug, its mode, and (if present) the trigger
PolicyThe policy type or policy names that applied
VerdictAllow, Would block, Block, or Error
ReferenceA content reference (no raw content)
LatencyEvaluation latency in milliseconds, when available

Verdicts:

  • Allow — the traffic passed.
  • Would block — the policy matched in observe (shadow) mode; the action was recorded but not blocked.
  • Block — the policy matched in enforce mode; the action was blocked.
  • Error — the check ran but could not reach a verdict (for example, a proxy error, network error, or timeout). These are recorded for audit completeness, not dropped.

A posture summary above the feed shows the count of each verdict type across your organization's recent decisions.

Agent actions

Each row is a single tool execution by an agent:

ColumnDescription
TimeWhen the action occurred
AgentThe agent slug and its identity
ActionThe action (tool) type
TargetA bounded target reference (no raw args)
OutcomeSuccess, Error, or Blocked

Both feeds support:

  • Search across agent, policy/action, and reference fields.
  • Team filter — scope both feeds to a single agent team (shown when your organization has more than one team).
  • Agent filter — scope to a single agent slug.
  • Verdict filter (decisions) / Outcome filter (actions).
  • Sortable columns — click a sortable header to sort by that column.

Click any row to open a detail drawer with the full (references-only) record.