Frequently Asked Questions
Common questions about Agent Inspector and how to use it effectively.
Getting Started
How many sessions do I need?
Minimum 5 sessions to start analysis, but meaningful behavioral patterns require 20+ sessions. More sessions = higher confidence in stability scores and outlier detection.
Is my data persisted or sent anywhere?
No. By default, all data is stored in-memory only. When you stop Agent Inspector, everything is cleared. No telemetry, no cloud uploads. You can optionally enable persistent storage to a local file for longer-term analysis.
Can I use this in CI/CD?
Yes! Run Agent Inspector as part of your test suite. Check the production gate status with
/agent-gate and fail the build if security findings block deployment. Perfect for pre-production validation.
Does it work with LangChain / other frameworks?
Yes! Any framework that calls Anthropic or OpenAI APIs works. Just point the
base_url to Agent Inspector's proxy (http://localhost:4000).
IDE Integration
How does IDE integration work?
Agent Inspector connects to your IDE (Cursor or Claude Code) via the MCP protocol. This gives coding agents a persistent memory layer — they remember security findings across sessions — and access to dynamic runtime data. Use slash commands like
/agent-scan, /agent-fix, and /agent-gate directly in your editor.
Do I need IDE integration to use Agent Inspector?
No. IDE integration is optional but recommended for the best experience. You can use Agent Inspector standalone with the proxy and dashboard. IDE integration adds static analysis, slash commands, and correlation capabilities.
What IDEs are supported?
Currently Cursor and Claude Code. Both use the MCP protocol for communication. More IDE integrations are planned.
Static vs Dynamic Analysis
What's the difference between static and dynamic analysis?
Static analysis scans your code for potential vulnerabilities without running it — like a security linter. Dynamic analysis monitors your agent during runtime to detect behavioral issues that only emerge during execution. Both are important: static catches code-level issues, dynamic catches behavioral anomalies.
How does correlation help prioritize issues?
Correlation connects static code findings with runtime evidence. A VALIDATED finding means the issue was both detected in code AND observed in runtime — fix immediately. A THEORETICAL finding was detected in code but never triggered in testing — lower priority. This helps you focus on real risks, not theoretical ones.
What security frameworks are supported?
Static analysis maps to OWASP LLM Top 10 categories. Reports also include SOC2 compliance mapping and CWE references. The dashboard shows coverage percentage against these frameworks.
Reports & Compliance
Can I generate reports for auditors?
Yes! Agent Inspector generates three report types: Security Assessment (technical details), Executive Summary (high-level for stakeholders), and Customer Due Diligence (for customers asking about your security posture). Reports include OWASP LLM coverage, remediation status, and audit trails.
What compliance frameworks are supported?
Reports map findings to OWASP LLM Top 10, SOC2 controls, CWE (Common Weakness Enumeration), and MITRE ATT&CK patterns. This helps security teams understand how agent risks map to their existing compliance requirements.
Technical
What's the difference from runtime monitoring?
Runtime monitoring observes production. Agent Inspector validates pre-production, catching issues before users see them. Think: CI/CD gates vs. production observability. Agent Inspector is the missing "security gate" in your agentic SDLC.
What ports does Agent Inspector use?
- Proxy:
localhost:4000— point your agent here - Dashboard:
localhost:7100— web UI - MCP endpoint:
localhost:7100/mcp— IDE connection
How do I fix a security finding?
Use
/agent-fix in your IDE to get an AI-powered, context-aware fix for any finding. Agent Inspector understands your code and suggests targeted remediation. You can then apply the fix, dismiss it (with reason), or mark it for review.
Still Have Questions?
Can't find what you're looking for? We're here to help!