Is this just a chatbot?
No. It is a search interface over a fixed, curated collection. The AI component extracts and synthesizes answers from documents in the collection. It does not have access to the open internet and does not generate information outside what is in the sources.
How is this different from asking ChatGPT?
ChatGPT draws from everything it was trained on and can state things with confidence that are wrong. Disaster Clippy only searches documents in the collection and shows you the source for every answer. If the answer looks wrong, you can read the original source yourself and judge it directly.
What is a collection?
A collection is the public-facing name for the source packs that power Disaster Clippy. In the app and on the site, you explore collections. Under the hood, those collections are built from portable source packs containing the source text, metadata, embeddings, and offline artifacts for one body of content. The hosted app comes with a preparedness collection, and local installs can load additional packs behind that collections layer.
Can I run this without internet access?
Yes. The local runtime supports fully offline use with a local language model via Ollama and local vector search. No external APIs required. It runs on a Raspberry Pi.
Can I use this for content other than disaster preparedness?
Yes. The engine is general-purpose. The hosted app demonstrates preparedness use, but the platform is designed for any curated body of knowledge: building codes, medical protocols, humanitarian reference archives, internal technical documentation.
Why is source creation local-only?
Scraping and ingestion are compute-heavy, domain-specific, and better suited to maintainers who understand their sources. Keeping creation local preserves the simplicity of the hosted app while giving full control to people who need it.
What language models does it support?
The hosted app uses a cloud API. Local installs support Ollama for running models on your own hardware, as well as API-based models if you supply your own keys. The model layer is configurable.
What vector databases does it support?
ChromaDB for local installs, Pinecone for cloud-backed deployments. The runtime selects based on configuration.
Why are there two repositories?
The public repo is the runtime: the app, ingestion tools, and source-building pipeline. That is everything you need to run the system and build on it. The private repo holds the .com site and internal maintainer docs.
Why is it called Disaster Clippy?
Clippy was Microsoft's attempt at a helpful assistant. It became famous for getting in the way. The name is a reminder of what this tool is trying not to be: an assistant that talks over its sources, gives you confidence without grounding, and makes itself the point. The goal is to stay out of the way and point you to the document. The name is also a placeholder — this is an early-stage project and the branding may change.