Configuring model settings of your chatbot

Instructions prompt

The instructions prompt is sent to the AI model along with the visitor’s message and your sources. It controls the chatbot’s personality, style and the type of responses it gives.

Pick one of the templates — AI Chatbot, Customer Support Agent, Sales Agent, Language Tutor, Coding Expert or Life Coach — or write your own. New chatbots start with the AI Chatbot template, which tells the model to answer from your sources, say when it does not have the information, and redirect unrelated requests.

The instructions tell your chatbot how to behave, define its personality and interaction style. You can:

• Adjust personality: For a casual, friendly bot, use instructions like “You are a friendly and casual AI assistant.”
• Handle unknown queries: Change what it says when the answer is not in your sources, for instance “I don’t have that information — you can reach our team at support@example.com.”
• Focus on specialized topics: Direct its expertise to specific fields, e.g. “You are an AI assistant who tutors children learning to code.”
• Set boundaries: Limit its response scope, such as “Do not share medical advice or information.”
• Response length and style: Tweak the general response style: “Be concise and concrete. Avoid jargon.”

This setting aligns the bot with your voice and values, enhances relevance, prevents misinformation, and avoids responding to topics beyond its knowledge base. Keep the instruction to answer only from the sources — it is the main protection against made-up answers.

AI model

Our chatbots are powered by OpenAI models. All current models are reasoning models: they “think” briefly before each answer, which makes them more accurate than the older chat models but adds a moment before the reply starts.

• GPT-5.6 Luna: The default. Fastest and cheapest, and a good fit for most support and FAQ chatbots.
• GPT-5.6 Terra: Balances intelligence and cost. A good step up when answers need more nuance.
• GPT-5.6 Sol: The GPT-5.6 frontier model, for the highest accuracy on complex questions.
• GPT-6 Astra: The most capable model, for the hardest work. Available on the Growth and Enterprise plans.

Terra, Sol and Astra require an upgraded plan; the model picker shows which ones your plan includes. Because these models reason for themselves, there is no “temperature” setting to tune — steer precision and tone through the instructions prompt instead.

Saving your changes

After changing any settings, click “Save” to apply them to your chatbot. Then ask a few questions in the Playground to check the effect.

Next steps

Once you’ve configured the model settings, adjust the look of your chatbot under Settings → Interface — name, bubble position, initial message, suggested messages, accent color and bubble picture — and then add the AI chatbot to your website.