Thriving 
PY HeteroAgent - LangGraph 1.x + FastAPI multi-agent customer service: deterministic routing, RAG, mandatory compliance, interrupt-based HITL, official MCP SDK, OTel GenAI semconv
HEALTH ? 71 / Stable
Thriving PY // readme
HeteroAgent · Intelligent Customer Service Multi-Agent System
A Supervisor-orchestrated customer service system built on LangGraph 1.x + FastAPI: deterministic routing → business Agents (RAG Q&A / tickets) → mandatory compliance review → aggregated reply. Built-in prompt injection pre-interception, PII detection and redaction, in-graph interrupt-based human-in-the-loop (HITL) approval, and hash-chain auditing.
Table of Contents
- Capabilities Overview
- Architecture
- Quick Start
- API
- Configuration
- Project Structure
- Quality Assurance
- Design Trade-offs
- Roadmap
Capabilities Overview
| Domain | Capability |
|---|---|
| Orchestration | LangGraph fixed topology (predictable/auditable), routing/tickets constrained by Pydantic schema, typed exception retries |
| Tools | Official MCP SDK (tooling/) + create_agent tool loop, each call produces an execute_tool span |
| Retrieval | FAISS vector + BM25 (single-character + two-character hybrid tokenization) hybrid retrieval + RRF fusion; scores normalized to [0,1]; supports Contextual Retrieval ingestion |
| Compliance… |