Project NOVA is a self-hosted AI assistant ecosystem designed to orchestrate 25+ specialized agents through n8n workflows and Model Context Protocol (MCP) servers. The system implements a hub-and-spoke architecture that routes user requests to domain-specific expert agents capable of controlling diverse applications ranging from knowledge bases and digital audio workstations to home automation systems and development tools.
The repository is written primarily in Shell and serves as a complete implementation guide for building a multi-agent AI system. It includes system prompts, Dockerfiles, and ready-to-import n8n workflows that enable users to deploy the entire ecosystem locally. The architecture centers on an intelligent router agent that analyzes incoming requests and conversation context to determine which specialized agent should handle each task, maintaining awareness of conversation history throughout interactions.
Project NOVA's specialized agents span multiple functional domains. Knowledge management agents interface with applications like TriliumNext Notes, Blinko, BookStack, Memos, Outline, SiYuan, Karakeep, Paperless-NGX, and ONLYOFFICE DocSpace. Development and repository management agents provide access to CLI execution, Forgejo, Gitea, and system-wide file search capabilities. Media and creative production agents control Ableton Live through Ableton Copilot, manage OBS Studio for streaming and recording, and interface with REAPER and other digital audio workstations. Additional agents handle AI and automation tasks including Flowise chatflows, Langfuse prompt management, browser automation via Puppeteer, RAGFlow for retrieval-augmented generation, and web content fetching. Home automation and monitoring agents connect to Home Assistant for smart device control and Prometheus for system metrics analysis.
The system supports two deployment approaches. The recommended Option A combines OpenWebUI as a modern chat interface with n8n as the workflow engine, providing conversation persistence and session management. Option B offers a basic n8n-only setup for users with simpler requirements. Both approaches require n8n version 1.88.0 or later, which introduced native MCP Client Tool node support, eliminating the need for community nodes. Users must provide Docker infrastructure for running containerized MCP servers and access to an LLM through either cloud APIs or a local Ollama instance.
According to GitGenius activity tracking, the repository shows minimal issue and pull request engagement, with a single tracked item exhibiting a median response latency of 508.9 hours. The primary contributor tracked is dujonwalker with one recorded event. The repository has been classified across numerous domains including AI development, machine learning, scalable architecture, automation, cloud computing, and computational tools, reflecting its broad applicability across different use cases.
The reference documentation is organized into comprehensive guides covering agent capabilities, the routing system, and categorized sub-agents. Users can implement scenarios ranging from voice-controlled home automation to knowledge base management, creative production workflows, development assistance, system monitoring, and content analysis. The setup process involves cloning the repository, choosing a deployment option, configuring either OpenWebUI with an n8n inlet filter or importing workflows directly into n8n, and connecting the desired applications that each agent is designed to control.
