Do I have to tell my agent not to make up answers?

No. The engine answers with what it knows (prompt, memories, catalog) and, when it doesn't know something, it says so and escalates. Instructions like "don't make things up" or "don't hallucinate" are unnecessary.

Does the agent pass the conversation to a person when needed?

Yes. If the customer asks for it or the agent can't solve it, the conversation goes to the team and is flagged in the inbox. You don't have to program it.

What is WhatsApp's 24-hour window?

WhatsApp only allows you to reply with free text within 24 hours after the customer's last message. Outside that window the agent uses approved templates, and the engine already respects that rule.

Help center

What TBit's AI agent already does without you configuring it

TBit's engine controls made-up answers, escalates to a person, respects WhatsApp's 24-hour window and remembers the conversation. What you shouldn't put in the prompt.

Published on July 18, 2026Updated on October 10, 20265 min read

What it does

TBit's engine comes out of the box with the behaviors that make a sales and service agent reliable. You don't have to ask for them in the prompt: they're already there.

What comes included

Control of made-up answers: the agent answers with what it knows (prompt, memories, catalog). When it doesn't know something, it says so and escalates. Instructions like "don't make things up" or "don't hallucinate" are unnecessary.
Escalation to a human: if the customer asks for it or the agent can't solve it, the conversation goes to the team and is flagged in the inbox.
WhatsApp format: short messages, no endless blocks of text, with the format WhatsApp supports. You don't have to explain what a good message looks like.
Meta's 24-hour window: WhatsApp only allows free replies within 24 hours after the customer's last message. Outside that window the engine uses approved templates. The agent already respects this rule.
Conversation memory: the agent remembers what was said in the conversation and the contact's data. It doesn't need the prompt to tell it "remember what they tell you".
Language: it replies in the customer's language.

Where you configure it

The agent's prompt lives in Training. There you define your business, personality and your own rules.

What NOT to put in the prompt

None of the above. The prompt is for what only your business knows: what you sell, how you talk, what policies you have, what makes your brand different. A short, specific prompt performs better than a long one full of plumbing.

Train with your own AI

Do you use ChatGPT or Claude? Connect them to your TBit agent and let them help you train it with your real data: on the platform, go to Agent Settings, the Connect your AI section, generate your token and add the connector https://api.tbit.app/mcp. Your assistant will be able to review the prompt, the memories and recent conversations, and suggest improvements.