ListeningKitWebsite index
Sitemap to indexed pages — routes, statuses, seeds, and the Sources UI.
This page describes an earlier design of ListeningKit, built on an in-browser demo, so parts of it do not match the app that runs today. For what works now, read the Guide.
Website index
The brand website is not a link on the record — it is indexed content. The sitemap resolves to pages, pages resolve to text, and text becomes the RAG namespace the agent quotes from.
Flow
- Onboarding (or the Brand tab) provides the website URL.
POST /brand/indexresolves the sitemap into a page list — or accepts an explicit{ urls[] }.- Each page is fetched and stored as a
BrandPage: url, title, headings, text,status,fetchedAt. - Statuses:
indexed(quotable),pending(queued),failed(retryable from the UI).
sources: [
{ url: 'https://acmeplumbing.com/services', title: 'Services', headings: ['…'], text: '…', status: 'indexed', fetchedAt: '…' }
]Mock rules
- Deterministic seed pages for the demo brand (homepage, services, about, contact) — no network in the hackathon build.
- Re-index replaces page text in place; failed rows stay visible with a retry action, never silently dropped.
- Replies cite sources via
sourceRefs: [{ url, excerpt }]on the query, so today's mock citations look exactly like tomorrow's RAG hits.
Sources UI
The Brand tab's Sources section lists every page: title, URL, status dot, fetched-at. Re-index site re-runs the pipeline with progress state; per-row retry clears failed rows one at a time.
The mock never fetches the live web. The live indexer swaps the fetch step only — page shape, statuses, and routes stay identical.