Cairn
An offline-first knowledge and education server: Wikipedia, courses, maps, local AI and RAG, with the internet unplugged.
- 01Command Center over Docker
- 02Offline by construction
- 03Local AI with retrieval
- 04Supply Depot
- 05Auto-updates with a dry run
- 06Open by default
Architecture Overview
A self-contained knowledge and education server, built as a derivative of Crosstalk Solutions' Project NOMAD under Apache-2.0 and rebranded so none of the upstream marks are reused. A management UI called the Command Center installs, configures and updates a set of containerised tools — offline Wikipedia and reference libraries through Kiwix, Khan Academy courses through Kolibri, regional maps through ProtoMaps, a data workbench through CyberChef, local notes, and a local AI chat with document upload and semantic search over Ollama and Qdrant. An internet connection is needed only during install and when you choose to pull new content; after that there is zero telemetry and nothing leaves the box.
How it works
Core mechanics, failure recovery paths, and system design decisions.
Command Center over Docker
A management UI and API orchestrate every tool as its own container, handling install, configuration and updates so a deployment is one script rather than a dozen compose files. A Debian-based host is the only requirement; no desktop environment, everything is reached through the browser.
Offline by construction
Connectivity is tested against 1.1.1.1 first and falls back to endpoints the app already contacts, so a network that blocks Cloudflare is not misread as offline. The check endpoint is overridable by env var or from the UI, and the whole stack runs with the link down.
Local AI with retrieval
Chat runs against a local Ollama model or any OpenAI-compatible server such as LM Studio, with uploaded documents chunked into Qdrant for semantic search. The model can also be pointed at a separate GPU host without moving the rest of the stack.
Supply Depot
A one-click catalogue of extra apps — PDF tools, file browser, e-book library, password manager — plus a path to run your own custom containers alongside the managed ones.
Auto-updates with a dry run
Opt-in updates install only minor and patch versions of the Command Center, inside a configurable window, after a cool-off, and only when pre-flight disk and queue checks pass. An Ace command runs the entire decision pipeline — window wrap, cool-off, prerelease, disk — without ever triggering an update, and the deterministic scenario suite is safe to wire into CI.
Open by default
There is no authentication: Cairn is meant to be reached without hurdles on a trusted LAN, and access is controlled at the network layer. It is explicitly not built to face the internet.
Engineering Highlights
- •Runs with the internet unplugged — install and content downloads are the only online steps
- •Zero built-in telemetry; nothing is sent anywhere at runtime
- •Connectivity check falls back past a blocked 1.1.1.1 instead of declaring the box offline
- •Auto-update logic has a full dry-run mode and a deterministic scenario suite for CI
- •Honest about scope: no auth by design, and not intended to be exposed to the internet
- •Derivative of Project NOMAD (Apache-2.0), with upstream names, domains and services removed per §6