KubeRiva OMS, an open-source AI-native order management system, launches with 7 sourcing strategies including an LLM-based adaptive mode and multi-tenant architecture.

KubeRiva OMS is a production-grade, async-first order management system built for mid-market e-commerce teams. It ships with Claude Haiku by default but allows users to bring their own LLM (a type of AI that understands and generates text). The system includes 7 sourcing strategies, with AI_ADAPTIVE using an LLM to score fulfillment nodes per order based on historical delivery rates, cost, and backorder data.

The system is deployed via pip/pipx (requiring Python 3.12+ and Docker Desktop) or Docker Compose, starting 9 services automatically. It features bidirectional connectors for Shopify and Amazon SP-API out of the box, a 15-state order lifecycle machine, multi-tenant isolation with per-tenant PostgreSQL databases, and role-based access control (RBAC) with 3 tiers.

An AI learning loop operates on a recurring schedule: hourly outcome labeling, nightly pattern discovery, and daily A/B experiment evaluation, with human approval required before rule changes are applied. A connector roadmap includes planned support for WooCommerce and Magento (v0.2), FedEx, UPS, and DHL (v0.3).