Controls & Audit

Manage guardrails, emergency agent controls, and account audit trails.

Keep agents safe and accountable. Review guardrail results, use agent controls during emergencies, and inspect the audit trail for account changes.

Protect agent activity

Review screened inputs and outputs. Stop an agent or all agents when immediate action is required.

Investigate changes

Trace configuration and access changes across the account.

At a glance

Area	Use it to	Start here
Guardrails	Review safety-rule evaluations	Guardrails
Agent Controls	Stop active agents during an emergency	Agent Controls
Audit Trails	Identify who changed what and when	Audit Trails

Use this area to answer three questions: what did guardrails catch, how do you stop an agent now, and who changed an account setting.

Guardrails

Guardrails screens agent executions against safety rules and records the result. Open Controls & Audit → Guardrails to see recent evaluations.

Guardrails, listing recent safety-rule evaluations.

This screen currently displays sample data — its title reads Guardrails (Mock). Treat the figures and rows as illustrative until it's connected to live guardrail evaluations.

You can narrow by agent and search by agent or guardrail. The page summarizes recent activity: Evaluations (24h), how many were Blocked and their share of traffic, the Pass Rate, and the number of Active Rules. Each row is one guardrail evaluation:

Field	What it shows
Timestamp	When the evaluation ran.
Agent	The agent whose input or output was screened.
Guardrail	The rule applied — for example, PII Filter, Toxicity, Topic Policy, Prompt Injection, Hallucination Guard, or Jailbreak Detection.
Type	Whether the guardrail ran on the Input (before the model) or the Output (after).
Result	Passed or Blocked.

Agent Controls

Agent Controls shows the runtime state of every agent and gives you emergency kill switches. Open Controls & Audit → Agent Controls.

Agent Controls, with emergency stop switches and per-agent runtime status.

At the top, Emergency controls gives you a Stop all agents button. Below it, you can search for an agent and see its state:

Field	What it shows
Agent	The agent's name.
Status	Its runtime state — Waiting, Completed, or Unknown.
Action	A per-agent control to stop that agent.

Stopping an agent aborts its in-flight threads immediately. You can start new runs again at any time, so the stop is not permanent — but work in progress when you stop it is lost. Use Stop all agents only in a genuine emergency.

Audit Trails

Audit Trails is the account's change log — every configuration and access change, in one place. Open Controls & Audit → Audit Trails.

Audit Trails, the account-wide change log.

You can narrow by workspace, by actor, and by date range, and search by actor, module, or action. Entries are listed most-recent first:

Field	What it shows
Timestamp	When the change happened.
Actor	Who made it (for example, [email protected]).
Module	The area affected — for example, Accounts, Marketplace, Users, Knowledge, or Agents.
Action	What happened — for example, Created, Updated, User Invited, or Generated.

Related

Control CenterEvaluation & QualityLogs & DataAccount Settings