Project NOMAD is a free, open-source server that puts Wikipedia, Khan Academy courses, offline maps and an optional local AI assistant on a computer you own. People on the same network use it in a browser, with no internet connection. Crosstalk Solutions makes it, and the latest stable release as of November 1, 2026 is v1.35.1.
That matters wherever the internet is slow, costly or gone: a rural school, a library branch, a field program or a training room. This piece walks through what Project NOMAD is, who makes it, what is in it, who it suits, what it takes to run, how install and updates work, its limits, and the alternatives worth comparing. Two later posts go deeper on hardware and storage, and on running it safely on a shared network.
The short version: Project NOMAD bundles several established open-source projects behind one web dashboard. Kiwix, Kolibri, Protomaps and Ollama do the heavy lifting. You install it with one script on Debian-based Linux, and it costs nothing under the Apache 2.0 license. The AI assistant is optional, and everything else runs without it on modest hardware. It has no login of its own, it is built for x86 Linux first, and help comes from a community. It suits small teams that want offline content without assembling each piece themselves.
What Project NOMAD is and who makes it
The project’s GitHub README calls it "a self-contained, offline-first knowledge and education server." You install it once while you are online and download the content you want. After that, it keeps working when the connection drops.
Under the hood, a management app called the Command Center runs each tool in its own Docker container. If containers are new to you, our guide to containers and Docker covers the basics. Other devices on the network reach the dashboard in a web browser.
Crosstalk Solutions builds it. According to Crosstalk Solutions’ own site, the company has designed and supported business networks and phone systems for about a decade. It also runs a large YouTube channel. Chris Sherwood signs the project’s story page as the creator of Project NOMAD. He writes there that he has never considered charging for it.
The project has drawn a lot of attention. As of November 1, 2026, the GitHub repository shows about 39.8k stars and 4.0k forks. Stars show interest. They do not tell you how well it will run in your building.
What Project NOMAD costs, and what it sends home
Project NOMAD is free. The project FAQ says it is free and open source under the Apache 2.0 license, with no subscriptions or paywalls. The project site says it relies on community support such as donations.
Apache 2.0 is a permissive open-source license. An organization can run it, change it and share it without paying a fee. Your real costs are the computer, the storage and staff time.
On privacy, the README says Project NOMAD has no built-in telemetry. It needs the internet only to install and to download optional content. Its connectivity check contacts Cloudflare’s 1.1.1.1 endpoint, and the README says you can change that in settings.
What is inside Project NOMAD
The README lists each feature and the upstream project behind it. Each project has its own site and community.
| Feature | Upstream project | What you get |
|---|---|---|
| Information Library | Kiwix | Offline Wikipedia, medical references, guides and ebooks |
| Education Platform | Kolibri | Khan Academy courses, progress tracking, multiple users |
| Offline Maps | Protomaps | Regional maps you download for offline viewing and search |
| AI Assistant (optional) | Ollama and Qdrant | Local chat with document upload and semantic search |
| Data Tools | CyberChef | Encoding, hashing, encryption and data analysis |
| Notes | FlatNotes | Local notes stored as markdown files |
| Supply Depot | Built in | One-click app catalog, plus your own Docker containers |
A few notes on the parts. Kiwix is a nonprofit, based in Switzerland, that packages Wikipedia and other sites for offline reading. Kolibri comes from Learning Equality, a US 501(c)(3) nonprofit, and is built for teaching without the internet. Protomaps builds its world map from OpenStreetMap and other open data and serves it from a single file. CyberChef runs entirely in the browser, and its source sits on GCHQ’s GitHub account.
The Supply Depot, according to the README, adds apps such as PDF tools, a file browser, an e-book library and a password manager. There is also a System Benchmark that scores your hardware against a community leaderboard.
The AI assistant is optional
The project site is clear on this point: if you do not want AI, skip it, and everything else still works. The FAQ says the same about the library.
When you do turn it on, Ollama runs open-weight models on the machine. Qdrant stores your uploaded documents so the assistant can search them before it answers. That pattern is called retrieval-augmented generation, or RAG. Our explainers on what Ollama is and how large language models work cover both ideas.
The README says Ollama is the preferred engine because it can download models. OpenAI-compatible servers such as LM Studio or llama.cpp also work. You can point the assistant at a model server on another machine on your network.
The AI is the part that needs real hardware. The README recommends a GPU-backed machine for it and notes that more video memory runs larger models. A sensible plan is to start without AI, see how people use the library and courses, and add the assistant later. Our guide to local AI models covers what to expect from them.
Who Project NOMAD suits
Project NOMAD fits teams that need the same reference content on many devices where the connection is weak. Four examples from our readers’ world:
- A rural school or library. Students and patrons join the local Wi-Fi and open Wikipedia or Khan Academy lessons in a browser. Teachers can track progress through Kolibri. No device pulls video over a thin rural line.
- A field program with poor connectivity. A health outreach or extension team carries one laptop with medical references and regional maps. Staff look things up between sites, with no signal needed.
- A disaster-response team’s go kit. A preloaded laptop holds maps of the service area, reference guides and the team’s own documents. It works when local networks are down. It fits inside a wider [disaster recovery plan](/it-infrastructure/what-is-disaster-recovery-planning/).
- A training room. One machine serves the same courses and references to a room of laptops. A workshop does not depend on the venue’s guest Wi-Fi.
It is a weaker fit for an office with good internet that mainly wants a private chatbot. A plain Ollama setup is simpler there. It also fits poorly where every user must sign in, for reasons covered under limits below.
Project NOMAD requirements in brief
The README gives two hardware profiles. It says the lists are for example and comparison only.
- Minimum: a 2 GHz dual-core processor, 4 GB of RAM, at least 5 GB of free disk, a Debian-based OS, and internet during install only.
- Optimal, for the AI tools: an AMD Ryzen 7 or Intel Core i7 or better, 32 GB of RAM, an NVIDIA RTX 3060 or AMD equivalent, and at least 250 GB of free disk, preferably SSD.
Storage drives most of the planning. The FAQ says the full English Wikipedia with images is about 125 GB, and that a full library fits on a 1 TB drive. The FAQ also says it is built for x86 PCs, not Raspberry Pi, and does not run natively on macOS. Windows users can follow a community-supported WSL2 guide.
Our next post covers hardware tiers and storage per content pack in detail. For now, note that local AI hardware prices rose sharply in 2026, which affects the AI tier most. Our checklist for buying an AI appliance also applies to a prebuilt NOMAD box.
How a Project NOMAD install works
The install guide assumes a dedicated x86 PC. You install Ubuntu Desktop on it, which erases the disk, and update it. Then you download and run one install script with administrator rights. The script sets up Docker, the containers and the service.
The guide says the script itself takes about a minute and the whole process takes about an hour, including downloads. When it finishes, you open the Command Center at localhost:8080 on the machine. Other devices use the machine’s IP address on port 8080.
An Easy Setup wizard then asks which apps you want, which map regions to download, and how much content to pull. Collections come in Essential, Standard and Comprehensive tiers, and a storage bar shows what fits. Experienced Docker users can skip the script and use the project’s Docker Compose template instead.
On versions, the README recommends Ubuntu 26.04 LTS and says Ubuntu 24.04 LTS and Debian 12 are also supported. The install guide says virtual machines are not officially supported.
How updates and the Supply Depot work
Automatic updates are opt-in. According to the README, they can cover the core software, installed apps and offline content on a schedule you control.
For its own updates, the Command Center can install minor and patch releases during a window you set. It waits through a cool-off period and runs only when pre-flight checks pass, such as enough free disk and no downloads in progress. Major versions always need a manual update. A helper script updates the Command Center and its dependencies only. Installed apps update through the dashboard.
The Supply Depot is where you add apps after setup. Pick from the catalog, or bring your own Docker container. Each added app is one more thing to patch and back up, so add only what someone will use.
Version status as of November 1, 2026
According to the project’s releases page, the latest stable release is v1.35.1, published October 5, 2026. It is a bug-fix release.
Version 1.35.0 came out on September 29. Its notes list image attachments for the AI, citations in document answers, an offline Translate app and map marker tools. Two release candidates came first, on September 13 and September 24. The release before that, v1.34.1, shipped on September 2.
No newer release or pre-release was listed as of November 1. The pattern so far is a release about every month, with release candidates before each minor version.
Limits of Project NOMAD to know before you commit
- Linux first. It targets x86 machines running Ubuntu or Debian. Windows is a community path through WSL2, and macOS and Raspberry Pi are not supported.
- Large downloads. Full Wikipedia with images is about 125 GB, per the FAQ, and each map region can be several gigabytes. Plan the first sync on a fast connection.
- AI needs real hardware. A small model on an older PC will be slow. The non-AI parts do not have this problem.
- Snapshots go stale. Offline content is a copy from the day you downloaded it. Schedule updates while you have a connection.
- Community project, no SLA. The FAQ points to Discord and GitHub for help. There is no support contract or response time.
Alternatives to Project NOMAD
Kiwix on its own. If you only need the library, Kiwix Server shares Wikipedia and other offline sites over a local network. Kiwix says it runs on Linux, FreeBSD, macOS and Windows, with Docker images, and on hardware as small as a Raspberry Pi Zero or a NAS. Kiwix also sells the Kiwix Hotspot, a Raspberry Pi 5 with its own Wi-Fi network. Its page lists it from CHF 260 as of November 2026.
Internet-in-a-Box. Internet-in-a-Box is free, open-source software for a Raspberry Pi or Linux PC that serves a local Wi-Fi learning hotspot. It pulls content packs from Kiwix and other libraries, adds OpenStreetMap maps, and can include Kolibri, Moodle or Nextcloud. Volunteers run it with schools, clinics and libraries. The Project NOMAD site itself points Raspberry Pi users there.
Kolibri on its own. If coursework with teacher tracking is the goal, install Kolibri directly. Its install docs cover Windows, Debian and Ubuntu, Raspberry Pi, macOS and Android.
Frequently Asked Questions
What is Project NOMAD?
A free, open-source, offline-first server from Crosstalk Solutions. It bundles Kiwix, Kolibri, offline maps, a few tools and an optional local AI assistant behind one web dashboard on your own computer.
Does Project NOMAD cost anything?
No. The project says it is free under the Apache 2.0 license, with no subscriptions or paywalls. You pay for the hardware, storage and staff time.
Do I have to use the AI assistant?
No. The project says the AI is optional and everything else works without it. You can add it later from the dashboard.
What hardware does Project NOMAD need?
The README’s minimum is a 2 GHz dual-core CPU, 4 GB of RAM and 5 GB of free disk. For AI, it suggests a Ryzen 7 or Core i7, 32 GB of RAM, an RTX 3060-class GPU and 250 GB or more of SSD storage.
Which operating systems does it support?
Debian-based Linux on x86. The README recommends Ubuntu 26.04 LTS and also supports Ubuntu 24.04 LTS and Debian 12. Windows has a community WSL2 guide. macOS and Raspberry Pi are not supported.
How do people connect to it?
In a web browser. On the machine itself, open localhost on port 8080. Other devices on the same network use the machine’s IP address on port 8080.
Does Project NOMAD have user logins?
No. The README says it has no authentication by design and should not be exposed to the internet. Keep it on a network you control.
What is the current version?
As of November 1, 2026, the latest stable release is v1.35.1 from October 5, 2026. Version 1.35.0 shipped September 29.
What are the main alternatives?
Kiwix Server or a Kiwix Hotspot for the library alone, Internet-in-a-Box for a Raspberry Pi learning hotspot, and Kolibri alone for coursework with teacher tracking.
