Hosted app
The hosted app requires nothing. Visit the site, ask a question, get a cited answer. The collection is maintained by the project. You can filter which sources to include but cannot add your own or change what is indexed.
This is the right choice if you want to use the preparedness collection as-is and do not need to bring your own data.
Local runtime
A local install runs the same interface on your own machine, pointed at your own collection. You control what goes in, which language model handles responses, and whether any external API calls are made at all. With a local model via Ollama, the system runs entirely offline.
Local installs work on a laptop, a home server, a Raspberry Pi, or an air-gapped node. This is the right choice if you want to bring your own data, run offline, or deploy for a specific context like a local organization or a field kit.
Advanced local admin
The admin toolkit extends the local runtime with source creation workflows: scraping websites, processing PDFs, indexing ZIM archives, running translation pipelines, and packaging the result into a portable pack. This is for people who want to build and curate knowledge, not just consume it.
The same local install that runs the chat interface also runs the admin panel. There is no separate installation — the admin is part of the public repository.
How they relate
A local deployment can sync with the hosted app's collection, run a completely independent collection, or mix both. Packs are the bridge - the same pack format works in both contexts. You can use the admin tools to build a pack locally and the local runtime to serve it, with no hosted dependency at all.