AI Assistant
The AI Assistant lets your chat widget answer visitors autonomously, using your own data, while always being ready to hand off to a human. It’s a separate, opt-in capability from the operator-assist tools (“Generate Reply” / “Improve Message”) — this one drives the conversation directly, on your team’s behalf.
What it does
When enabled for an application, the AI Assistant:
- Reads the visitor’s message and replies directly in the widget — no operator has to be online.
- Answers using your own data, through admin-configured tools (an HTTP endpoint or a read-only database query — see Configuration). It never invents facts: it only states what a tool actually returned in that conversation.
- Hands off to a human the moment it can’t help — see Escalation & Handoff.
Escalate, never fabricate
This is the core guarantee behind the assistant: if it doesn’t have a grounded answer, it stops and connects the visitor to your team instead of guessing.
When it escalates
Every AI-handled conversation ends one of two ways: a grounded answer, or a clean handoff to a human. The assistant escalates whenever:
- It doesn’t have enough information to answer (no tool, or a tool returned nothing useful).
- The visitor asks for a person, or clicks the “Talk to a human” button that’s always visible in the widget header.
- Something technical fails (a tool call errors out, the request times out).
- Your monthly AI usage quota is exhausted.
There’s never dead air — the visitor is either helped or queued for an operator, and your team sees it land in the queue in real time.
AI transparency
Visitors always know they’re talking to an AI:
- Every AI reply is labeled (“AI Assistant” by default, configurable) with a distinct avatar in the widget.
- The first message in a new conversation discloses that an AI assistant is available and that a person is always reachable.
- A persistent “Talk to a human” button lives in the chat header at all times — the visitor is never stuck with only the AI.
This is a deliberate design choice, not just good UX: several jurisdictions (e.g. California’s B.O.T. Act, the EU AI Act) require clear disclosure when a user is interacting with an automated system.