AI Website Chat Agent
Step 2 of 5

Import knowledge

Add public pages, sitemap, feed, file, or Q&A sources. Preview what was found before anything is approved for answers.

Preview sourceOnly fetched, content-indexed sources are used for grounded AI answers.

Production setup journey

Add website → import knowledge → set instructions → install widget → test chat / go live

Each step now follows the real launch sequence: knowledge and instructions come before answer testing.

  1. Add websiteConfirm the public domain visitors will use.Pending
  2. Import knowledgeAdd public pages or a sitemap before testing answers.Pending
  3. Set instructionsTell the assistant how to answer, escalate, and stay on brand.Pending
  4. Install widgetSet appearance, copy the public snippet, and check detection.Pending
  5. Test chat / Go liveRun a safe preview, then confirm launch readiness.Pending

Knowledge health

Start here: what can the assistant actually use right now, what is only a stored URL, and what is blocked/backing off.

Status
Loading knowledge health…

Only rows marked as fetched/content-indexed are usable grounded KB. URL-only rows are visible so you can decide whether to crawl, refetch, upload, paste, or archive them.

Knowledge → Quick Addendum

Quick Addenda

Fill a gap in the assistant’s knowledge quickly. Add the missing information, give one example question, then add it to the knowledge base.

How Quick Addenda work
  1. Write and save a draft. Nothing is used in answers yet.
  2. Preview retrieval. Check the exact questions and page scope that will match.
  3. Approve and publish. Only users with publish permission can make it available to answers.

Only published Addenda can answer. Paused, expired, superseded or conflicted items are excluded.

Saved addenda

Choose an item to review or edit it. Status and version are shown on each row.

Loading addenda…
Add missing knowledgeTell the assistant what it should know and give one example question.
Ready to fill a knowledge gapEnter the missing information and one question a customer might ask.
What should the assistant know?
Write the information as you want the assistant to explain it.
Scope, source & timingAdvanced — narrow where the fact applies or set review and expiry dates.
Preview or advanced actions
More lifecycle actions
Retrieval test results
Retrieval results appear here.
Version history and rollback

Usable KB items first

Grouped source batches with content-fetched pages ahead of URL-only/blocked work. Expand a source to see the exact pages the assistant can or cannot use.

Loading knowledge sources…

Add or repair knowledge

Use this after reviewing KB health: choose a source, preview what we found, then approve it before the assistant can use it.

Choose a KB source

What do you want the assistant to learn from?

Pick the closest option first. It pre-fills the same import form below, keeps preview/approval in place, and gives advanced users control over source type and refresh.

Advanced and blocked-site options

Choose a source type above, or fill the import details manually.

Upload TXT, Markdown, HTML, CSV/TSV, JSON/JSONL, XML, YAML, LOG, RTF, PDF, DOC/DOCX, or XLS/XLSX. Text-like files load into a draft preview automatically; for PDF/Office files, paste extracted text into the content box for now. Uploading the same filename creates a new version that supersedes the older KB file after approval.

Nothing is added to live answers until you approve the preview. Same-name file uploads supersede the older version only after approval.

Find website pages instead

If you prefer website discovery, paste a home/support page. We fetch that one public page, list same-site links, and let you choose exactly which pages to queue — no broad crawl.