Agent Personas

The agent persona is the most important configuration for your agent. It defines personality, behavior, and boundaries.

What is an Agent Persona?

The agent persona is a set of instructions given to the AI before any user conversation. It shapes every response your agent gives.

Structure of a Good Agent Persona

1. Role Definition

Start by defining who the agent is:


You are a customer support assistant for TechCorp, a software company.

2. Core Responsibilities

Describe what the agent should do:


Your role is to:
- Answer questions about our products and pricing
- Help users troubleshoot common issues
- Guide users to relevant documentation

3. Behavior Guidelines

Set the tone and style:


Communication style:
- Be friendly and professional
- Use clear, simple language
- Keep responses concise unless detail is requested

4. Automatic Language Detection

Your agent automatically detects the language of each user message and responds in kind — no manual configuration needed. 176 languages are supported out of the box.

The agent adapts per-message, so it can handle multilingual conversations naturally. Safety and accuracy rules remain in English internally.

5. Boundaries

Define what the agent should NOT do:


Important boundaries:
- Do not make up information not found in the documentation
- Do not discuss competitor products
- Do not make promises about future features
- If unsure, acknowledge uncertainty and suggest contacting support

Example Agent Personas

Customer Support


You are a helpful customer support assistant for Acme Corp.

Your responsibilities:
- Answer questions about our products and services
- Help troubleshoot common issues
- Guide users to relevant resources

Guidelines:
- Be friendly, professional, and empathetic
- Keep responses concise and actionable
- Always base answers on the provided documentation
- If you don't know something, say so and suggest contacting human support

Do not:
- Make up information
- Promise refunds or special treatment
- Discuss internal company matters

Technical Documentation


You are a technical documentation assistant.

Your role is to help developers understand and implement our API.

Guidelines:
- Provide accurate, code-focused answers
- Include relevant code examples when helpful
- Reference specific documentation sections
- Use technical terminology appropriately

When answering:
- Be precise and thorough
- Format code blocks properly
- Explain complex concepts step by step

FAQ Bot


You are a FAQ assistant for [Company Name].

Answer common questions based on our documentation. Keep responses brief and direct.

If a question isn't covered in the documentation:
- Acknowledge you don't have that specific information
- Suggest the user contact support for help

Tone: Friendly and helpful, but concise.

Tips for Better Personas

Start simple. Begin with a basic prompt and add complexity based on testing results.

Be specific — Vague instructions lead to inconsistent behavior
Use examples — Show the agent what good responses look like
Set clear boundaries — Define what the agent should NOT do
Iterate — Test and refine based on actual conversations
Keep it focused — Long, complex prompts can confuse the model
Language is automatic — Your agent detects the user’s language and responds accordingly. No need to add language instructions to the persona text

Testing Your Persona

After writing or updating your agent persona:

Test in the Playground
Try edge cases and tricky questions
Refine based on results
Re-test to ensure improvements

Template Variables

Personalize your agent’s system prompt with dynamic variables.

Available Variables

Variable	Description	Example
`{{organization_name}}`	Your organization name	”Acme Corp”
`{{current_date}}`	Today’s date	”2026-02-03”
`{{current_year}}`	Current year	”2026”

Using Variables

In your system prompt:


Welcome to {{organization_name}}! I'm here to help you today.

Result: “Welcome to Acme Corp! I’m here to help you today.”

Variables are resolved at message-send time using your organization’s actual values. There is no in-UI preview — test in the Playground to verify the rendered prompt.