Import flows
Beamdesk ingests public help centers, individual URLs, CSV exports, and uploaded files. Each import creates source records, article candidates, embeddings, crawl metadata, and a review state. Sources stay visible in the Sources panel so operators can see what the AI is allowed to cite.
Sitemap
Use for public help centers and documentation sites with many pages.
URL
Use for one high-value page, policy, changelog, or temporary launch note.
CSV
Use for historical FAQs, macros, canned replies, or exported help center articles.
File
Use for PDFs, DOCX files, internal runbooks, and product one-pagers.
POST /api/knowledge/ingest
{
"type": "sitemap",
"url": "https://example.com/sitemap.xml",
"collection": "public-docs",
"auto_publish": false
}Watchtower draft loop
Low-confidence queries are not just failures. Beamdesk Watchtower groups them into knowledge gaps, suggests article drafts, and links each draft to the unresolved tickets that motivated it. This creates a review queue from real customer language instead of guesses.
Open /kb/drafts to approve, edit, merge, or reject AI-suggested drafts. Approved drafts become retrievable knowledge immediately after indexing.
Wikilinks and backlinks
Use wikilinks to connect related articles inside Beamdesk: [[Article Name]]. The editor resolves the title, stores a structured link, and shows backlinks on the target article.
Backlinks help reviewers find outdated dependency chains. If a pricing article changes, backlinks reveal onboarding guides, cancellation answers, and sales notes that mention the same policy.
Refund policy summary
Customers on monthly plans can cancel from Settings.
For exceptions, follow [[Refund exception workflow]].
For annual plan changes, see [[Annual billing changes]].Sources panel
The Sources panel shows schedule, last crawl, document count, dead-link detection, crawl errors, and manual recrawl controls. Use it before debugging answer quality: many "bad AI" issues are stale sources, broken redirects, or duplicate pages with conflicting policy.
Manual recrawl is useful after pricing changes, launch notes, outage updates, or website migrations. Scheduled crawls should be frequent for public docs and slower for stable internal runbooks.
Freshness scoring
Every article receives a freshness score based on age, source crawl health, linked ticket volume, recent edits, and detected contradictions. Stale articles are still searchable, but Beamdesk lowers confidence and may request review before auto-resolving a ticket with them.
Review stale articles when they influence high-volume intents, reference changing policies, or appear in low-confidence answer traces. Archive articles that are obsolete rather than leaving them unpublished and duplicated.
Approving drafts
The draft queue at /kb/drafts is designed for fast editorial review. Each draft shows source tickets, proposed title, answer body, citations, suggested tags, confidence impact, and merge suggestions.
Before approving, check that the draft answers one intent, links to related articles, avoids private customer data, and names the owner who can keep it current.