Feasibility & architecture for making the Technical Support Assistant the first-line agent inside Amphenol's existing website live chat — with seamless escalation to a human engineer.
Amphenol uses LiveChat.com (developer platform now branded “Text”, platform.text.com). Our self-hosted FastAPI multi-agent backend plugs in as a native Bot Agent — it answers from our own LLM (not ChatBot.com's no-code builder) and, on request or low confidence, hands the live conversation to a human engineer via transfer_chat with full context retained.
Exactly the model Amphenol asked for: the AI answers instantly, a human steps in the moment it's needed.
Every new chat fires a webhook; our backend runs the multi-agent LLM (grounded in FitMate data + verified sources) and replies as the bot agent — instantly, 24/7. Same engine as the standalone chatbot POC, just a different transport.
On an explicit “talk to a human” or low model confidence, the bot calls transfer_chat to reassign the live conversation to a human engineer or group — no restart, full context retained.
incoming_event fires for any event in a chat (including other agents/bots) — the backend must filter by author / visibility so the bot never replies to itself.
A handoff only succeeds if a human is available and accepting chats. We implement a fallback path (e.g. “our team is offline — leave your email”) when no agent is free.
The customer keeps LiveChat's native widget — no website rebuild. Only the agent side is our custom bot.
| Need | Mechanism |
|---|---|
| Register the bot | Configuration API → create_bot_agent (name, status, group priority, webhooks) |
| Receive customer messages | Webhooks: incoming_chat (new chat), incoming_event (each message) — HTTP POST, return 200, retries on failure |
| Reply as the bot | Agent Chat API send_event (message event) with X-Author-Id = bot identity |
| Escalate to a human | transfer_chat → target type agent or group; confirmed via chat_transferred push |
| Authentication | OAuth 2.1. Authorization Code Grant for a private server-side app; Personal Access Token for simpler service access; Client Credentials for a headless bot identity |
| Customer UI | LiveChat's native widget stays (single script tag). Only the agent-side bot is custom — no site rebuild |
The same bot-first → human-handoff pattern is supported by the other majors — so the approach isn't locked to one vendor.
| Vendor | Custom bot + handoff | Mechanism |
|---|---|---|
| LiveChat.com (in use) | Yes | Bot Agent · Agent Chat API · transfer_chat |
| Intercom | Yes | Fin AI + Canvas Kit for custom bots |
| Zendesk | Yes | Sunshine Conversations bot-to-agent handoff |
| Salesforce | Yes | Agentforce / Omni-Channel |
| Freshchat | Yes | Bot-conversation assignment to agents |