Connecting a knowledge box#
Administration lives under Manage and is passcode-protected. A portal needs a knowledge box connected before search and answers work. Connect an existing box by binding it, or create and provision a new one from within the app - the portal configures the box (its taxonomy, graph, agents and suggested questions) for the domain you describe.
Bindings are held server-side; the credentials never reach the browser. Every call to the content platform is made from the server.
Adding content#
Add content into the connected box several ways:
- Upload documents (PDFs and other files) directly.
- Add a link to a web page for the box to crawl and ingest.
- Add text as a resource.
Ingestion is asynchronous. When the box is busy processing recent changes it applies back-pressure; the portal waits and retries within bounds, and tells you honestly when the box is too busy to accept more right now rather than failing silently.
Ingesting and syncing web sources#
Point the portal at a source site and it discovers the linkable pages so you can ingest them as a set. Sources can be re-synced on a schedule so the corpus keeps up with a site that changes, without anyone re-adding pages by hand.
Corpus health#
The corpus-health view scans the connected content for problems - failed ingests, documents whose text extracted thin or empty, bot-challenge pages that slipped in, and raw untitled entries. From here you can re-ingest what needs fixing and permanently purge the genuinely broken entries (a narrowly-scoped, confirmed delete), keeping the corpus that users see genuinely clean.