nosia: Self-Host AI RAG and MCP on Your Data

Overview

Nosia is a self-hosted AI RAG and MCP platform built with Ruby on Rails and PostgreSQL with pgvector. It lets teams query their own documents and connect AI models to external tools and data sources, while keeping the application and data on infrastructure they control. OpenAI-compatible endpoints let existing clients connect to Nosia.

The project is aimed at people who want a packaged platform rather than assembling separate document search, chat, and MCP components. Its Docker Compose setup targets macOS, Debian, Ubuntu, and Windows, and the README describes using Docker Hub AI models by default, with configuration for other compatible model services.

Key Features

• Index PDFs, text files, websites, and question-and-answer pairs for semantic search.
• Add retrieved document context to chat completions through configurable RAG.
• Connect MCP servers over stdio, SSE, or HTTP so chats can use external tools and resources.
• Expose an OpenAI-compatible API with streaming responses.
• Provide account-based data isolation and a web interface for documents, chats, and integrations.
• Run the application and supporting services with Docker Compose.
• Extend chat behavior with custom LLM-driven and Ruby-based skills.

Use Cases

• A small team can ask questions over internal documents without sending those documents to a hosted knowledge platform.
• Developers can connect existing OpenAI-compatible clients to a self-hosted endpoint for document-grounded chat.
• Teams can give an assistant access to selected calendars, file stores, or messaging tools through MCP servers.
• Administrators can deploy a shared, account-isolated AI workspace when they are prepared to operate its Docker-based services.

Getting Started

On macOS or Linux, the README provides a one-command installer:

curl -fsSL https://get.nosia.ai | sh

Then start the services with docker compose up -d and visit https://nosia.localhost. The README and deployment guide cover configuration, production TLS, and upgrades: README and deployment guide.

Alternatives

• MaxKB: MaxKB centers on building knowledge-based agents and workflows, while Nosia combines document RAG and MCP connections with an OpenAI-compatible API.
• rag-agent-platform: rag-agent-platform uses Java and Next.js for configurable retrieval, agent, and MCP workflows; Nosia is a Rails and PostgreSQL platform with an OpenAI-compatible API.

Considerations

• This is an application stack, not a single executable: Docker Compose runs the web application alongside services including PostgreSQL, model, embedding, and background-job components.
• The README's default setup uses Docker Hub AI models. Model choice, embedding dimensions, and compatible AI service configuration need to match the deployment; changing embedding dimensions can require rebuilding existing embeddings.
• The default local endpoint uses a self-signed certificate. Production deployments need proper SSL configuration, and the README notes that Windows users may need to add nosia.localhost to their hosts file.
• The project is active, with regular commits and releases, and contributions from multiple people. However, the measured record shows no issues closed in the recent 90-day period, so teams should evaluate their support expectations and review open issues.
• The MIT license permits broad use, subject to its license terms and notices.