In practice
Ask a question. Read the answer. Check the references.
Disaster Clippy keeps the answer and the supporting sources side by side so you can move from summary to verification without leaving the search flow.
Open-source / Self-hostable / Offline-capable
A self-hostable knowledge system that shows its sources.
Disaster Clippy lets people ask practical questions, inspect the source behind each answer, and carry the same pattern into their own local deployment. The hosted app is a demo of the model. The real goal is to help people run a trustworthy version on their own hardware and their own data.
Demo, docs, GitHub
Try the demo, read the platform docs, then install from GitHub when you are ready to run your own deployment. A simpler launcher path can come later, but GitHub is the path today.
1. Try the demo
2. Read the platform docs
3. Clone from GitHub
Demo behavior
We lost power after a storm. What should we do first?
Start with safety, water, refrigeration, lighting, and backup communication. Then inspect the strongest matching references in the collection before acting.
Why someone would deploy this
The preparedness demo is one example of the pattern. The same runtime supports building codes, humanitarian archives, internal manuals, PDFs, transcripts, and Kiwix ZIM archives in local or offline deployments. The point is not to depend on a hosted preparedness chatbot. It is to run a trustworthy, inspectable knowledge system on your own hardware, with the sources you choose.
Building codes, field manuals, local archives, internal references, and preparedness material can all use the same runtime once the library is assembled around them.
Hosted demo
See the interaction before you install it
The live demo lets you test the search pattern on a preparedness collection before you decide whether to run the same approach on your own hardware.
Grounded answers
The source stays in view
Answers are meant to stay tied to the collection that produced them, so you can inspect the references instead of taking a model on faith.
See the interaction, then run it yourself
The hosted app shows how grounded answers and visible references feel in practice. The docs and repo let you run that same pattern on your own hardware and your own data.
Open the demoNot a black-box chatbot
Disaster Clippy answers from the collection in front of you. If the answer looks wrong, you can inspect the underlying source instead of trusting hidden training data.
See the trust modelBuilt to be run locally
The long-term value is not just using a hosted app. It is being able to run a version yourself, keep the knowledge bounded, and carry it into offline or constrained environments.
See deployment optionsBuilt for your own data
The preparedness collection is the example. The same engine can support your own documents, archives, PDFs, transcripts, or offline libraries once you are ready to build a deployment around them.
See the platform pathHow this is different
- Not open-web search: answers come from a bounded, curated collection you can inspect -- not whatever ranks or whatever a model was trained on.
- Not cloud-only: the same system runs on your own hardware, offline, on a Raspberry Pi, with no external API calls required.
- Not just document chat: collections are versioned, portable artifacts with a validation and source governance layer -- not one-off workspaces.
- Not a black box: every answer cites the source it came from. If the answer looks wrong, you can read the original document yourself.
From demo to deployment
Start with the demo if you want to test the interaction. Go to About if you want the trust model. Explore Collections if you want to see what the preparedness demo is built from. Use the technical docs if you want to understand the runtime, ingestion path, and how to run your own version.
Read enough to understand the model, then head to the repo when you are ready to install and run it yourself. Today that path is GitHub; a simpler launcher can come later.