What you can ingest
The ingestion engine handles most common document and web sources without custom code. Point it at a MediaWiki site, a folder of PDFs, a static HTML site, a Substack newsletter, or a ZIM archive and it produces a searchable, source-cited collection.
Supported source types include MediaWiki sites (Appropedia, Fandom wikis, or any MediaWiki instance), PDF files and folders with section-aware chunking, static HTML sites discovered via sitemap, Substack newsletter archives, and ZIM files including offline Wikipedia snapshots and video transcript archives.
What the admin panel does
The local admin panel runs alongside the app on your own machine. It gives you a job queue for long-running ingestion work, source management with per-source metadata and licensing, a validation layer that checks document quality before indexing, and a 3D visualisation of the knowledge graph to help you understand what is in the collection and how it connects.
Translation and language pack workflows let you index non-English sources while keeping the search layer in normalized English, with optional per-language retrieval for localized deployments.
Why it stays local
Scraping, transcript acquisition, and indexing are compute-intensive, domain-specific, and require decisions about source quality and licensing that are better made by the person who knows the material. Keeping these workflows local preserves the simplicity of the hosted app and gives the operator full control over what goes in.
The hosted app runs on a fixed, curated collection. The local admin is how that collection is built and maintained, and how you build your own.
What a deployment looks like
A local deployment runs the same chat interface you use at the hosted app, pointed at your own data. It can run on a laptop, a local server, or a Raspberry Pi. With a local language model via Ollama, it runs entirely without external API calls.
The same collection can be packaged as a portable pack and carried offline, synced to a cloud vector database for a hosted deployment, or shared with others running compatible installs.
Getting started
The runtime and all local tooling are in the public repository. The README covers installation, configuration, and the first steps for bringing your own data.
View the repository on GitHub