Skip to main content

Quick Start

Quick Start gets you from an empty organization to cited answers in the Admin Console, then optionally to a live website widget.

Introduction​

In Qefro, knowledge is scoped to a workspace. Customer AI (widget / WhatsApp) and Employee AI (Internal Portal) both bind to workspaces you configure in the Admin Console.

Why it exists​

Validating retrieval quality before connecting Business Tools or public channels avoids shipping an inaccurate assistant.

Concepts​

  • Document — PDF, DOCX, Markdown, TXT, or crawled pages; chunked and indexed with hybrid BM25 + vector retrieval
  • Citation — answers include source references; the model is instructed to refuse when no relevant source exists
  • Widget token — publishable embed key from Widget in the Admin Console (rotatable)

Architecture​

Workflow​

First productive workspace

  1. Create workspace — Admin Console → Workspaces (e.g. Customer Support).
  2. Add knowledge — Documents → upload files or start a website crawl.
  3. Set instructions — Define tone, languages, and refusal behavior for the assistant.
  4. Test in console — Ask in-scope and out-of-scope questions; check citations.
  5. Embed widget (optional) — Widget page → copy script with data-token and data-workspace-id.

Code examples​

Website embed (values come from Admin Console → Widget):

<script
  src="https://cdn.qefro.com/widget.js"
  data-token="YOUR_WIDGET_TOKEN"
  data-endpoint="https://api.qefro.com"
  data-theme="auto"
  data-position="bottom-right"
  data-workspace-id="YOUR_WORKSPACE_ID">
</script>

Best practices​

  • Start with 5–20 high-quality docs, not a full drive dump
  • Keep customer-facing and employee knowledge in separate workspaces
  • Test refusal: ask something outside the corpus

Security notes​

FAQ​

How long until uploads are searchable?
Usually seconds to a few minutes depending on size and OCR.
Is chat WebSocket or HTTP?
The widget prefers WebSocket at /ws/chat?token=… with HTTP SSE fallback at POST /api/v1/widget/chat.