Rosmarium COS: AI-Native Content Orchestration System

#typescript #fastify #graphql-yoga #drizzle #postgresql #redis #bullmq #python
Visit Project → View on GitHub
Rosmarium COS: AI-Native Content Orchestration System

Rosmarium COS: The AI-Native Content Orchestration System

Rosmarium COS is an open-source, AI-native headless content orchestration system built with a pragmatic split-stack architecture. With the recent release of Version 2.0, it has evolved into a fully-fledged enterprise-grade platform that seamlessly integrates with both modern frontend applications and asynchronous background AI workers.

V2 Features & Architecture

The 2.0 release introduces powerful new orchestration capabilities designed for scale, collaboration, and developer autonomy:

  • 🌿 Content Branching: A Git-like branching model for content. Teams can isolate, stage, and merge massive content revisions safely without affecting production data.
  • 👁️ Visual Live Preview: Integrated visual previewing leveraging dual-channel WebSockets, giving editors real-time, component-level feedback directly within the headless UI.
  • 🛡️ Enterprise RBAC: Granular Role-Based Access Control and multi-tenant isolation built on Node’s AsyncLocalStorage and PostgreSQL search paths.
  • 🧱 Structured Block Documents: A modern approach to rich text, moving away from static HTML strings to strongly-typed JSON block documents, paving the way for advanced frontend rendering and AI manipulation.

The AI Intelligence Pipeline

Traditional CMS platforms treat AI as an afterthought. Rosmarium bridges this gap natively with its dedicated Python-based AI worker pool:

  • 🕸️ Semantic Knowledge Graph: AI workers continuously process content to extract entities and build a traversable semantic graph utilizing spaCy and NetworkX.
  • 🔍 Hybrid Vector Search: Leveraging pgvector inside PostgreSQL, Rosmarium delivers lightning-fast semantic search combined with RRF (Reciprocal Rank Fusion) and traditional BM25.
  • 🤖 Generative Workflows: Integrated LlamaIndex streaming enables real-time auto-tagging, summarization, and generative content workflows.

Technology Stack

  • TypeScript & Fastify: The high-performance core API server, providing completely type-safe runtime validation via Zod.
  • GraphQL (Pothos & Yoga): A fully automatic, code-first GraphQL layer offering deep querying, robust connection pagination, and WebSocket subscriptions.
  • Drizzle ORM & PostgreSQL: A type-safe, schema-driven persistence layer.
  • BullMQ & Redis: A robust asynchronous messaging layer powering the transactional outbox webhook system and internal event pipelines.
  • Python (FastAPI): A dedicated, heavily-typed backend AI worker handling heavy GPU/inference workloads securely.