How NaviBot works

NaviBot has three main layers:

1. Website widget - the visible assistant and chat overlay.
2. Client cabinet - where the client configures sites, widget behavior, OpenAI, Knowledge sources, integrations, and analytics.
3. Backend brain - the 1Desk-style system that stores conversations, dispatches AI replies, keeps operators informed, and owns server-side data.

Runtime flow

1. The website loads https://navi.1it.pro/widget.js.
2. The widget creates or resumes a visitor session.
3. It collects safe page context: title, visible sections, links, FAQ-like text, and relevant page structure.
4. The visitor sends a message.
5. The backend stores the message in the conversation.
6. The dispatcher checks whether the inbox has an AI agent integration.
7. The OpenAI adapter builds a prompt from:
   • site prompt
   • behavior rules
   • live page context
   • selected RAG documents
8. The reply is saved and emitted back to the widget.

Product boundary

NaviBot is the face on the website. The backend remains the brain.

Area	Owner
Widget UI	NaviBot
Character and motion	NaviBot
Page context scan	NaviBot widget
Conversations	1Desk-style backend
OpenAI routing	Backend integration layer
Knowledge sources	NaviBot cabinet + backend RAG
Human handoff	1Desk-style inbox/operator flow
Analytics	NaviBot events + backend storage

Required setup

Before launch, prepare:

• target website domain
• website inbox in the cabinet
• OpenAI account or managed OPENAI_API_KEY
• one initial page map or page context target
• widget embed snippet
• privacy / consent text
• first set of Knowledge sources

Common issue

If the widget accepts messages but does not answer, check whether the website inbox has an OpenAI agent integration and whether an OpenAI key exists.