AI Agent Orchestration Starter Kit: Production-ready multi-agent AI kit with memory & HITL

Multi-agent AI orchestration with a supervisor that decomposes tasks and delegates to specialists — plus persistent memory (Redis + ChromaDB), human-in-the-loop approval for sensitive actions, and full OpenTelemetry observability with cost tracking and replay. Built and battle-tested on a real production system — including fixing a runaway loop that generated 5,000+ duplicate DB rows before it was caught. One-command Docker setup, ~2 minutes to running.

Hey Product Hunt! 👋 I built this after shipping a full multi-agent orchestration system (LangGraph, memory, human-in-the-loop approvals, full tracing) as a personal project — then realized most of the real pain was in the infrastructure, not the agent logic itself. Along the way I hit and fixed some real production bugs: a runaway approval loop that generated 5,000+ duplicate rows, a silent checkpoint failure, and a replay identifier mismatch. Those fixes are already baked into this kit, so you don't have to hit them yourself. This is for developers building AI agent products who don't want to spend 2-3 months on the orchestration/memory/observability layer before they can even start on their actual product. Would love feedback — especially from anyone who's built something similar and hit different walls!

Quick update for anyone following this launch — got the first 2 sales since launching here, and the first payout already came through. Really appreciate everyone who checked it out, upvoted, or left feedback. If you tried it out and hit any friction, genuinely want to hear about it — still actively improving this based on real usage.