Skip to main content

Cleanlab Logs

The Logs page gives engineers, product owners, and technical stakeholders full observability into your user’s and AI’s behavior, allowing you to view each interaction, Cleanlab’s guardrails and evaluations, and the interventions made by Cleanlab.

This comprehensive audit interface provides you with complete visibility into every AI interaction, including detailed evaluation results, guardrail results, and Cleanlab’s protective interventions. For AI-safety remediation workflows, see the dedicated Issues page.

Cleanlab Logs Interface

Detailed Log Information​

When troubleshooting a specific User-AI interaction, click on any log entry to open a detailed slideout view with comprehensive information:

The detailed view provides rich information for troubleshooting, including:

  • Retrieved Context: See exactly what information was retrieved from your knowledge base
  • Guardrail & Evaluation Scores: View detailed scores and pass/fail statuses for all configured guardrails and evaluations
  • Full Conversation History: Access the complete conversation context leading up to the interaction
  • Custom Metadata: Review any arbitrary metadata logged with the interaction (user location, entry points, etc.)
  • AI Response Details: Examine the complete AI response and any modifications made by Cleanlab
  • Cleanlab Interventions: See how Cleanlab handled the response:
    • Original LLM response was passed through
    • Original LLM response was guardrailed with a safe fallback
    • Original LLM response was replaced by an expert answer inputted as an Expert Answer Remediation in Cleanlab AI Platform

Detailed Log Slideout

Filtering and Analysis​

Filter by Guardrail Status​

  • View only interactions where specific guardrails were triggered
  • Analyze blocked responses and SME interventions
  • Track guardrail effectiveness over time

Filter by Evaluation Performance​

  • Identify patterns in AI response quality
  • Focus on specific evaluation criteria
  • Compare performance across different time periods

Combined Analysis​

  • Correlate guardrail triggers with evaluation scores
  • Understand which types of responses are most likely to be blocked
  • Optimize both safety and performance metrics

Integration with Guardrails & Evaluations Pages​

From the Logs page, you can:

  • Navigate to Guardrails: Click on guardrail names to view detailed configuration
  • Navigate to Evaluations: Click on evaluation names to see configuration and trends
  • Test Configurations: Use the test feature to validate guardrail/evaluation logic
  • Update Settings: Modify thresholds and criteria based on log analysis

For detailed configuration of guardrails and evaluations, see the dedicated Guardrails & Evaluations page.