Skip to main content

Agent Creation Logs

The Logs page provides a complete audit trail of all agent creation activities, including successful creations, pending approvals, denials, and errors. Use it to monitor agent behavior, investigate issues, and satisfy compliance requirements.

Accessing Logs

Navigate to System Admin > Agent Self-Service > Logs.

Stats Overview

At the top of the page, four summary cards provide a quick snapshot of creation activity:

CardDescription
Total AttemptsThe total number of agent creation requests
SuccessfulAgents that were successfully created
Pending ApprovalRequests currently awaiting administrator review
Denied/ErrorsCombined count of denied requests and failed attempts

These stats update when the page loads and when you click Refresh.

Filtering Logs

Click the Filters toggle to expand the filter panel. The following filters are available:

FilterDescription
Start DateShow logs from this date onward (default: 30 days ago)
End DateShow logs up to this date (default: today)
OutcomeFilter by result type (see outcome types below)
SearchFilter by agent name

Filters are applied automatically as you change them.

Outcome Types

OutcomeDescription
CreatedAgent was successfully created and is now in the system
Pending ApprovalRequest is awaiting administrator review on the Approvals page
DeniedRequest was rejected by an administrator or by the safety system
ErrorCreation failed due to a technical error (validation, quota, etc.)
DeactivatedAgent was automatically deactivated after its expiration period
DeletedAgent was automatically deleted after its expiration period

Logs Table

The main table displays the following columns:

ColumnDescription
TimestampDate and time of the creation attempt (UTC)
Creating AgentThe agent that initiated the creation request
Created AgentThe name of the agent that was created (or the proposed name for pending/denied requests)
Risk ScoreAI-generated danger score as a percentage, color-coded by severity
OutcomeStatus badge showing the result
DetailsClick to open the detail panel

Click any row to view full details about that creation attempt.

Log Detail Panel

Clicking a log entry opens a slide-out panel on the right side of the screen with comprehensive information:

Overview

  • Log ID - Unique identifier for this log entry
  • Timestamp - When the attempt occurred (UTC)
  • Outcome - Result with color-coded badge
  • Danger Score - AI risk assessment percentage

Agents

  • Creating Agent - Which agent submitted the creation request
  • Created Agent - The resulting agent (if creation succeeded)
  • Template Used - Which template was referenced (if any)

Creation Parameters

The full JSON payload of the creation request, including:

  • Proposed agent name and display name
  • System instructions
  • Requested tools
  • Execution settings (timeouts, tool call limits)
  • Expiration configuration

Validation Result

The AI safety analysis output, including the reasoning behind the danger score.

Error

If the creation failed, the error message is displayed here. Common errors include:

  • "Agent creation by agents is not enabled for this tenant" - Feature is disabled in Settings
  • "Rate limit exceeded" - The creating agent has hit its per-hour, per-day, or total quota
  • "Tool not available" - The agent requested a tool it doesn't have access to
  • "Invalid agent name format" - The proposed name doesn't match the required pattern

Downloading Logs

Click the download icon in the detail panel header to export an individual log entry as a JSON file. This is useful for:

  • Sharing log details with other administrators
  • Archiving for compliance purposes
  • Investigating issues offline

Pagination

Logs are loaded in batches of 50. If more logs are available, a Load more logs link appears at the bottom of the table. Click it to load the next batch.

Best Practices

  • Review logs weekly - Regular review helps identify patterns and potential issues early
  • Investigate errors - Error outcomes often indicate configuration problems that should be addressed
  • Monitor risk scores - A trend of increasing risk scores may indicate agents are pushing boundaries
  • Export important logs - Download logs for significant events (first creation, denied requests, errors) for your records
  • Use date filters for investigations - When troubleshooting, narrow the date range to the relevant period
  • Cross-reference with approvals - If a log shows "pending_approval", check the Approvals page to ensure it gets reviewed