Conversational search over your own data, online, offline, and locally.
Disaster Clippy is an offline-capable AI search system for any curated body of knowledge. It combines a hosted demo, a local runtime, and advanced local admin/source tools around a shared catalog and source-pack model.
It is not just a chat app and not just an admin tool. The project is meant to support four related surfaces:
- hosted demo app
- product and documentation site
- downloadable local runtime
- downloadable advanced local admin toolkit
Disaster preparedness is the first example deployment. The engine underneath is general purpose.
The core idea is simple: most AI systems hide what they know and you have to trust them. This one makes the knowledge layer visible, swappable, and verifiable. Every answer cites its source. You control what goes in. If something looks wrong, you can check it yourself.
That approach works for any curated body of knowledge. The same engine has been used for:
- Building codes — searchable local and regional code references for construction and compliance
- Pandemic preparedness — flu response protocols and field guidance for organizational planning
- Humanitarian reference — offline-capable search across Appropedia, Akvo, and similar open knowledge bases
The source pack model is what makes this portable. On the public site and in the app, people explore collections through a catalog. Under the hood, a source pack is the discrete, versioned bundle of knowledge — HTML backups, ZIM archives, PDFs, embeddings, and metadata — that can be installed locally, carried offline, or swapped out entirely. The runtime does not care what the pack contains.
If you are a developer thinking about building something similar for your own domain, the architecture is designed with that in mind. See docs/clippy-core-extraction.md for the direction the portable core is heading, and docs/template-deployments.md for the planned plug-and-play deployment path.
Users ask questions in plain language and get grounded answers from curated source packs.
Example questions:
- "What are the main code requirements for rooftop solar in this jurisdiction?"
- "Show me the strongest references in this internal operations manual"
- "What does this humanitarian guide say about water storage?"
- "Find practical how-to material across this archive"
The system is designed so the same general experience can work:
- in a hosted demo app
- on a laptop or local server
- on a Raspberry Pi or air-gapped node
The hosted demo is the easiest way to experience Disaster Clippy.
It is intended for:
- people exploring the system for the first time
- teams validating a source pack catalog before self-hosting
- users who want search/chat without running admin tooling
The local runtime is for people who want resilience, control, and their own data.
It is intended for:
- offline setups
- LAN/Raspberry Pi deployments
- organizations or individuals carrying selected source packs locally
The advanced admin layer is for power users and maintainers.
It includes:
- source creation and validation
- scraping and indexing
- translation workflows
- video processing workflows
- future OCR support for scanned PDFs
This is intentionally local-first. Disaster Clippy is not currently aiming to turn source creation into a public hosted workflow.
Source packs are the main unit of knowledge distribution.
A source pack may include:
- HTML backups
- ZIM archives
- PDFs
- transcript and translation artifacts
- metadata and vector indexes
This lets users:
- understand what content they are using
- choose which source packs to install or sync
- carry the same pack choices between hosted and local use
On the public site and in the app, these are presented as collections in a catalog. In the runtime and tooling, source packs are the portable unit underneath.
Current hosted demo:
https://app.disasterclippy.com
The current local install path is still the public GitHub repo and python app.py.
The wrapper/installer path is planned, but it is not the default install flow yet.
# Clone the repository
git clone https://github.com/xyver/disaster-clippy-public.git
cd disaster-clippy-public
# Install dependencies
pip install -r requirements.txt
# Configure local settings
cp .env.example .env
# Start the app
python app.pyThen open:
http://localhost:8000
# Local embeddings
EMBEDDING_MODE=local python app.py
# Optional local LLM via Ollama
ollama pull llama3.2:3bcurl -X POST "https://app.disasterclippy.com/api/v1/chat" \
-H "Content-Type: application/json" \
-d '{"message": "How do I purify water?"}'| File | Use Case |
|---|---|
requirements.txt |
Full local runtime and advanced admin tooling |
requirements-cloud.txt |
Lean hosted/runtime deployment |
Use the full install when you want:
- local ChromaDB
- source tools
- translation models
- video processing
- ZIM/PDF handling
- offline-oriented workflows
Use the cloud install when you want:
- a smaller hosted deployment
- chat/search without the heavy local processing stack
- source-pack-aware search across a curated catalog
- natural language search across curated collections
- grounded responses with source attribution
- source filtering
- multi-turn conversation
- hosted or offline-capable usage paths
- local source storage
- offline semantic search
- local model support
- source-pack installation from the catalog or local files
- source creation and validation
- background jobs and checkpointing
- translation and language pack workflows
- video transcript acquisition and processing
- future OCR preprocessing for scanned PDFs
disaster-clippy-public/
|-- app.py
|-- admin/
|-- cli/
|-- docs/
|-- offline_tools/
|-- static/
|-- templates/
The broader workspace direction is now:
disaster-clippy/
+-- disaster-clippy-public/
+-- disaster-clippy-private/
Where:
disaster-clippy-public/is the core runtime and local tooling repodisaster-clippy-private/is the.comsite and future control-plane shell
docs/deployment.md- runtime model, public/private split, hosted vs localdocs/distribution-alignment.md- what recent DaedalMap distribution updates imply for Clippydocs/architecture.md- system architecture and runtime layersdocs/source-tools.md- source creation and ingestion toolsdocs/language-packs.md- translation systemdocs/video_processing.md- video and transcript pipelinedocs/clippy-core-extraction.md- portable core architecture target
MIT for project code.
Content sources keep their own original licenses.