Overview
The Instructions tab in Agent Builder → Settings → Instructions lets you customise how each assistant greets visitors and shapes its personality. Everything set here is appended to the assistant’s system prompt — it does not replace the built-in prompt, only extends it.
Select the assistant you want to configure from the dropdown at the top of the tab. Settings are saved per assistant.
Welcome Message
The first message displayed when a chat session opens. By default this is pre-filled using the assistant’s name and description. Edit it to set a custom greeting.
Clear the field and save to restore the assistant’s built-in default welcome message.
Response Style
Controls the assistant’s tone and verbosity. The selected style is automatically appended to the system prompt.
- Balanced — default, general-purpose responses
- Concise — shorter, to-the-point answers
- Detailed — thorough, comprehensive responses
- Technical — precise and technical language
- Friendly — warm, conversational tone
Suggested Prompts
Up to 4 clickable prompt pills displayed below the welcome message. When a visitor clicks a prompt, it is sent to the assistant automatically. Leave a field blank to hide that prompt.
Use suggested prompts to highlight the most common questions for that assistant — they help visitors get started without needing to know what to ask.
Persona Notes
Additional context injected into the system prompt on every conversation for this assistant. Use it to share business context, tone of voice guidelines, or specific behaviour instructions — without overwriting the core system prompt.
Examples of what to include:
- Where this assistant is deployed and what it is for
- The business name, location, and owner’s name
- Tone of voice rules (e.g. always be warm, never use jargon)
- What to do or avoid in specific situations
Knowledge
Reference information this assistant can draw on during conversations. This is injected into the system prompt as a knowledge file — separate from any trained vector store data.
Ideal for:
- Business hours, location, and contact details
- Frequently asked questions and answers
- Product categories, pricing, and policies
- Any concise reference data the assistant should always have access to
For larger content — PDFs, full website pages, or extensive documentation — use the Knowledge page to upload files and train the vector store. Trained data is available to all assistants via retrieval-augmented generation (RAG).