What is Project NOMAD.?
Project NOMAD (N.O.M.A.D.) is a local offline AI server for Linux. It runs large language models, reference libraries, and maps on your own hardware. It also bundles a full offline education platform into the same install. NOMAD stands for Node for Offline Media, Archives, and Data. The entire system runs on Docker and operates on your local network with no telemetry and no data sent to external servers. The offline AI is free and open source but you must have a powerful machine to run it (min 32GB RAM, RTX3060, 250GB SSD) and some technical knowledge for the installation. Not compatible with Win/Mac/Rasperry PI/Virtual Machines.
Features & Benefits
- Local AI Chat with RAG — run large language models offline via Ollama and query your own documents through semantic search via Qdrant
- Offline Wikipedia & Reference Library — browse the full Wikipedia archive, medical references, repair guides, and ebooks locally via Kiwix
- Offline Education Platform — stream Khan Academy courses and K-12 lessons with local progress tracking for multiple users via Kolibri
- Offline Maps & Navigation — download regional OpenStreetMap data and plan routes without cell service via ProtoMaps
- Local Data Tools — run encryption, encoding, hashing, and data analysis offline via CyberChef
- Local Notes — write and store markdown notes on your own machine via FlatNotes
- Setup Wizard — pick which offline tools to install through a guided first-time setup
- OpenAI-compatible API support — point the local AI at LM Studio or llama.cpp instead of Ollama
- Docker-based Command Center — manage every component through a single local UI and API
- Zero Telemetry — no data collection and no outbound calls once installation is done
Real-World Applications
Sending documents to a cloud LLM means your data leaves your machine. NOMAD runs a local LLM with RAG so you can query and analyze documents on your own hardware. The model reads your uploaded files directly during the chat. That matters when you’re working with proprietary code, internal docs, or client data that shouldn’t sit on a third-party server.
Homelab users who already run self-hosted services can drop NOMAD in as an offline AI node on their network. It installs on any Debian system and supports GPU inference for larger models. Community builds range from old desktops to custom rigs. Benchmark scores run from 10 to 95 depending on the hardware.
If you’re already running LM Studio or llama.cpp, you can point NOMAD at that setup instead of installing Ollama. That way you keep your existing AI stack and just add NOMAD’s content bundle on top — maps, reference library, and education platform. The Docker setup means you can add or remove pieces without touching the rest of the system.
Linux users who want a full offline reference stack on a local machine can get it here. The Information Library covers Wikipedia, medical references, and repair guides through Kiwix. ProtoMaps handles offline maps with no cell service needed. The local LLM keeps working no matter what the connection looks like.
Self-hosters who want full control over their data can swap out cloud AI, notes, and reference tools for a single NOMAD install. CyberChef handles encryption and encoding locally. FlatNotes covers markdown notes. Everything runs on your own box with no outside calls.
