Agent Self-Service Settings
Configure how AI agents can create new agents and SMEs within your organization. The Settings page controls the master feature toggle, approval workflow, rate limits, and safety thresholds.
Accessing Settings
Navigate to System Admin > Agent Self-Service > Settings.
Enable Agent Creation Tool
The master toggle that enables or disables the entire feature for your tenant. When disabled, all agent creation requests are rejected with the error: "Agent creation by agents is not enabled for this tenant."
To enable the feature:
- Toggle Enable Agent Creation Tool to on
- Configure the remaining settings as described below
- Click Save Changes
After enabling the feature, you must also assign the create_agent and/or create_sme tools to the agents that should be allowed to create new agents. An agent can only grant tools it already has access to.
Require Approval Workflow
When enabled, agent creation requests are placed in a review queue instead of being executed immediately. An administrator must approve or deny each request on the Approvals page before the agent is created.
This is the recommended setting for production environments.
With approval disabled, agents can create new agents immediately without human review. This is not recommended for production environments.
Create Agents as Inactive by Default
When enabled, newly created agents start in an inactive state and must be manually activated by an administrator. This provides an additional layer of control even when the approval workflow is disabled.
This setting is enabled by default.
Rate Limits
Rate limits prevent any single agent from creating an excessive number of agents. Three limits work together to provide layered protection:
| Setting | Description | Default | Range |
|---|---|---|---|
| Max Per Hour | Maximum agents any single agent can create per hour | 5 | 1-50 |
| Max Per Day | Maximum agents any single agent can create per day | 20 | 1-200 |
| Total Limit | Maximum total agent-created agents in your tenant | 100 | 1-1000 |
When a rate limit is exceeded, the creation request is rejected and logged with an appropriate error message.
Choosing Rate Limits
- Low volume / conservative: 2 per hour, 5 per day, 25 total
- Default: 5 per hour, 20 per day, 100 total
- High volume / autonomous: 10 per hour, 50 per day, 500 total
Rate limits apply per creating agent, not globally. If you have 3 agents with the creation tool, each gets its own per-hour and per-day quotas. The total limit applies across all creating agents within the tenant.
Instruction Safety Threshold
Every agent creation request includes instructions (a system prompt) that are analyzed by AI for potential risks. The safety threshold determines when a creation request gets flagged for additional review.
How It Works
- When an agent submits a creation request, the proposed instructions are sent to an AI model for safety analysis
- The AI evaluates the instructions for risks such as data exfiltration, privilege escalation, self-replication, or harmful behavior
- A danger score between 0% and 100% is assigned
- If the score exceeds the configured threshold, the request is flagged
Risk Score Levels
| Score Range | Level | Color | Meaning |
|---|---|---|---|
| 0-29% | Low | Green | Instructions appear safe |
| 30-69% | Moderate | Yellow | Review carefully before approving |
| 70-100% | High | Red | Automatically flagged, likely unsafe |
Adjusting the Threshold
Use the slider to set the threshold between 0% and 100%:
- Lower threshold (e.g., 10%) = Stricter, flags more requests for review
- Higher threshold (e.g., 70%) = More permissive, flags fewer requests
- Default: 30% = Balanced approach suitable for most organizations
Examples
| Proposed Instructions | Typical Score | Why |
|---|---|---|
| "Research market trends and summarize findings" | ~5% | Benign research task |
| "Access all user emails and forward to external address" | ~85% | Data exfiltration risk |
| "Create additional agents to replicate yourself" | ~90% | Self-replication risk |
| "Handle customer support inquiries using knowledge base" | ~10% | Standard support role |
Saving Changes
After modifying any settings, click Save Changes at the top of the page. The button only appears when you have unsaved changes. You can also click Cancel to revert to the previously saved values or Refresh to reload the current settings from the server.