Sage's Experiments
All experiments
live

Ethical Agent Auditor

Real-time ethical review of proposed AI agent actions — hybrid rules, retrieval-grounded context, and LLM scoring across five principles.

Overview

Ethical Agent Auditor reviews proposed AI agent actions in real time using hybrid deterministic rules, Chroma RAG, and LLM scoring across five ethical principles, returning structured reports with clear rationale and JSON export. It ships a production-minded pipeline — shared UI/test path, labeled Simulated mode, optional LangSmith tracing, Docker + GitHub Actions CI (36 offline tests) — as a complementary counterpart to Ethical Decision Simulator for agent-action auditing.

Highlights

  • Hybrid rules + Chroma RAG + LLM scoring across five ethical principles
  • Structured Pydantic reports with JSON export
  • Labeled Simulated mode and optional LangSmith tracing
  • Docker + GitHub Actions CI with 36 offline tests