AgentAudit

Visual Decision Trajectory & Compliance Reporting

📋 Load Sample Log
Privacy Verified: 100% Client-Side
✓ Ready for Export
0
Total Steps
0
Decision Nodes
0
Risk Flags

Decision Trajectory

Audit Detail Log

Step Type Timestamp Content

Frequently Asked Questions

What Agent logs are supported?
AgentAudit supports standard JSON logs and plain-text execution traces from frameworks like LangChain, AutoGPT, LlamaIndex, and custom Agentic loops. The parser identifies common patterns including Thought:, Reasoning:, Action:, Tool Use:, Observation:, Result:, Final Answer:, and Response:.
Is my log data secure?
Yes. All parsing, visualization, and PDF generation happen locally in your browser. No data is ever uploaded, transmitted, or stored on any external server. The entire audit process is performed client-side, ensuring complete privacy and data sovereignty.
How does the Trajectory Mapping work?
The engine uses heuristic pattern matching to identify 'Thoughts', 'Actions', and 'Observations', linking them into a vertical timeline for easy auditing. Risk keywords like error, failed, unauthorized, and refused automatically trigger visual warning indicators, highlighting potential compliance issues.
Can I use the PDF report for compliance?
Yes, the "Export Report" feature generates a high-contrast, professional-grade document suitable for archival and compliance review. The report includes a verification timestamp, local audit ID, and structured presentation of the audit trail that meets enterprise governance standards.
AgentAudit is a specialized utility for parsing Agentic workflows. It identifies Thought-Action-Observation loops to create a visual audit trail. It is designed for developers auditing LLM behavior and governance teams ensuring AI safety. Key features include: visual trajectory mapping, automated risk detection, local PDF generation, and privacy-first processing. Supported formats: LangChain logs, AutoGPT traces, LlamaIndex logs, custom agent logs in JSON or plain text.