Governed Agentic RAG + Tool Use with LangGraph, LangChain, MCP, evaluation and human approval.
User/API → Policy pre-check → LangGraph supervisor → Retrieval/MCP tools → Evidence grading → Approval boundary → Synthesis → Trace + evaluation
The executable repository includes a deterministic no-key orchestration/evaluation mode. Provider-backed LLM quality metrics are intentionally not claimed until executed and published.
Stateful orchestration, typed tool protocol, side-effect governance, reproducible evaluation, observability contracts, API/container deployment, and CI/CD.