project-nomad (N.O.M.A.D, for Node for Offline Media, Archives, and Data) is a self-contained, offline-first knowledge and education server. It bundles local AI, an offline encyclopedia, courseware, maps, and data tools behind a single web-based command center, so a box with no internet can still answer questions, teach, and navigate. The honest framing is that it is not one piece of software but an orchestration layer: a Docker stack plus a content-pack manager that wires together well-known open-source components. That assembly, packaged for non-experts, is the value.
What is in the box
The built-in capabilities, per the README, are a curated bundle:
- Local AI chat with a knowledge base: Ollama for inference plus Qdrant for vector retrieval (RAG), with document upload.
- Offline reference: Kiwix-served offline Wikipedia, medical references, ebooks, and survival guides.
- Education: Kolibri with Khan Academy courses, progress tracking, and multi-user support.
- Offline maps: ProtoMaps with downloadable regional maps, search, and navigation.
- Data tools: CyberChef for encoding, hashing, and analysis, plus FlatNotes for local markdown notes.
The management layer itself is TypeScript/Node, with MySQL and Redis behind it and a self-updater sidecar.
Install and hardware
sudo apt-get update && sudo apt-get install -y curl && \
curl -fsSL https://raw.githubusercontent.com/Crosstalk-Solutions/project-nomad/refs/heads/main/install/install_nomad.sh \
-o install_nomad.sh && sudo bash install_nomad.sh
# then open http://localhost:8080The installer pulls dependencies, creates the Docker containers, and brings up a browser command center where a setup wizard picks content packs. Hardware expectations are notable: a minimal install (management framework only) wants a dual-core CPU, 4GB RAM, and 5GB storage, while running a useful local LLM realistically wants something like a Ryzen 7 or i7, 32GB RAM, an NVIDIA RTX 3060 or better, and 250GB+ SSD. This is deliberately not a low-power Raspberry Pi project; it assumes GPU acceleration.
When it fits, and when it does not
It fits emergency preparedness, off-grid education (a self-contained classroom), and anyone who wants a private, local knowledge-and-AI box that keeps working when the network does not. It fits less well if you want a minimal low-watt device (the Pi-class projects below are lighter), or if you want a single tightly-integrated app rather than an orchestration of separate services you will occasionally have to debug.
How it compares
| Project | Shape | Stars (2026-06) |
|---|---|---|
| Crosstalk-Solutions/project-nomad | Offline AI + knowledge server, GPU-oriented | ~30k |
| iiab/iiab (Internet-in-a-Box) | Offline knowledge on Raspberry Pi | ~2k |
| kiwix/kiwix-android | Offline Wikipedia reader | ~1k |
| meshtastic/firmware | Off-grid mesh communication | ~8k |
Internet-in-a-Box is the closest peer (offline knowledge plus Kiwix on modest hardware), but it targets low-power devices for education in the developing world, while project-nomad assumes a GPU and adds local LLM chat. Kiwix is the offline-library engine project-nomad uses; Meshtastic addresses the communications angle the community keeps requesting.
Gotchas from the issue tracker
The orchestration nature shows in the open issues:
- An axios version bump broke the TypeScript build, blocking PRs until resolved (#987).
- An Ollama
api/embedendpoint change broke embeddings, the kind of upstream drift an orchestration layer inherits (#959). - Requests to upgrade the Kolibri container off an old image (#982) and recurring asks for grid-down communication and Apple Silicon inference show the community pushing scope outward.
The pattern: because project-nomad stitches together Ollama, Qdrant, Kiwix, Kolibri, and more, its bugs are often upstream version drift and container maintenance rather than its own logic. Expect to manage updates.
A note on scope
Worth being clear-eyed: the positioning leans heavily toward prepper and off-grid survival use, not pure AI tooling, and the community frequently requests survival-specific add-ons. If you came for an AI knowledge base, that is in here, but it ships inside a broader offline-survival vision. Evaluate it as a packaged solution and content manager, not as a single application.
FAQ
Is project-nomad free? Yes. project-nomad is Apache-2.0 licensed and open-source; you supply your own hardware and the bundled components are open-source too.
What hardware does project-nomad need? The management framework runs on a dual-core CPU with 4GB RAM and 5GB storage, but a useful local LLM realistically wants a GPU like an RTX 3060 or better, 32GB RAM, and a 250GB+ SSD. It is not a low-power Pi project.
Does project-nomad work offline? Yes, that is the point: once content packs are downloaded, the local AI chat, offline Wikipedia, maps, and courseware all run with no internet.
How do I install project-nomad? Run its install_nomad.sh script with sudo on a Debian-based system, then open the browser command center at http://localhost:8080 and pick content packs.
Related reading
For the agent and local-AI side that powers the chat layer, see how local models are used in tools like anomalyco/opencode; for document ingestion into a knowledge base, microsoft/markitdown is a useful companion.