Setting the AI agent's behavior

AI Agent → Behavior controls who the agent is, how it writes, and how careful it is before sending. These settings shape every reply, so you rarely need to spell them out in instructions.

AI Bot Replies

The main switch. While it is off, the agent never answers and visitors always wait for your team.

Identity

Agent name is shown as the sender in the widget and in your inbox. It defaults to your project name followed by "AI". Give it a name that fits your brand, and be honest that it is an AI: customers respond better when they know.

Style

Tone of voice: Friendly, Professional, Formal or Casual.

Answer length: Brief, Balanced or Detailed. Brief suits a product with simple questions. Detailed suits technical products where a complete answer saves a follow-up.

Instructions

Optional free text that tells the agent how to behave: what to emphasise, what to avoid, how to greet. Keep it short and concrete. Good instructions read like a note to a new support hire:

You are the support agent for ACME. Greet the visitor by name when you know it. Use plain language. Never promise refunds; point to the refund policy instead.

Instructions do not add knowledge. If the agent needs to know a fact, put it in an article, a training source, or a snippet.

Accuracy

Double-check answers before sending makes the agent re-read each reply against your knowledge and score it for being backed by your content and for actually answering the question. A reply that scores below 70 is rewritten once. If the rewrite still falls short, the conversation goes to a person instead of the customer receiving a shaky answer. Leave this on unless speed matters more than precision; it adds a moment to each reply.