Deploying Agents

Learn how to move agents from draft to active status, dispatch work items, and monitor agent sessions through CleverGuard.

Key Concepts

Agent lifecycle

Agents follow a defined lifecycle: draft → active. Draft agents exist in the system but cannot process work. Activating an agent makes it available to receive and execute work items. If the Agent Review is enabled, the agent must pass agent review before activation.

CleverGuard sessions

When an agent is dispatched to work on a task, a CleverGuard sessionis created. This session enforces the agent's guardrail profile — tool allowlists, content scanning, budget limits, and drift detection. Every action the agent takes is logged and auditable through the SOC (Security Operations Console).

Drift detection

Clevername tracks a version hash for each agent, computed from its model, system prompt, expertise, and environment. If any of these change after review governance, drift is detected and the agent is blocked from deployment until it is re-governed.

Step-by-Step Guide

Activate the agent

From the AI Company dashboard, click on a draft agent to open its detail panel. Click the Activate button. If the Agent Review is enabled for your organization, you will be prompted to submit the agent for agent review instead.

Agent detail panel showing a draft agent with an Activate button highlighted

Draft agents show an Activate button. If agent review is required, you'll be directed to submit for review first.

Review governance (if required)

If your organization has the Agent Review enabled, activating a draft agent requires submitting the 43-question security questionnaire. Reviewers then review and vote. Once the minimum governance threshold is met, the agent transitions to active status automatically.

Note

See Agent Review → Submitting for Review for a detailed walkthrough of the submission process.

Dispatch a work item

With the agent active, you can dispatch work items to it. Navigate to the agent's detail page or use the Quick Actions menu on the dashboard. Click Dispatch Work, fill in the task description and parameters, and submit.

Dispatch Work modal showing task description field, parameter inputs, and a Submit button

Describe the task and set any parameters. The work item is assigned to the agent and a CleverGuard session is created.

Monitor the session

Once dispatched, a CleverGuard session is created and the agent begins processing. You can monitor the session from:

The agent's detail page — shows active and recent sessions.
The SOC console at /dashboard/security/soc — shows all agent sessions across the organization with full audit trails.
The Work Items tab — shows task status (in_progress, completed, failed).

SOC console showing an active CleverGuard session with action log, guardrail status, and tool usage timeline

The SOC console provides real-time visibility into what your agents are doing, including every tool call and content scan.

Handle escalations

If an agent encounters a situation that requires human intervention — such as a high-risk action matching a guardrail pattern — it triggers an escalation. Escalations appear in the dashboard stat cards and can be routed through SignedApproval for secure human-in-the-loop approval.

Important

Agents with escalation-triggering guardrail patterns will pause execution until a human approves or rejects the action. Ensure someone is available to handle escalations during agent operating hours.

Creating Agents Work Items Agent Review Overview Monitoring & Costs