> ## Documentation Index
> Fetch the complete documentation index at: https://docs.conduit.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Persona

> Set an agent's character, tone and hard limits in the Persona tab, and know what belongs in Knowledge or Skills instead.

Persona is where you tell an agent who it is and how it should sound. It is one of the tabs on an agent, alongside Activity, Tools, Skills and Settings, and it holds two things: **Instructions** and **Guardrails**.

Instructions set the voice. Guardrails set the lines the agent must not cross. Everything else about the agent lives elsewhere, so Persona stays about character rather than facts or process.

## Where to find it

Open **Agents** from the workspace sidebar, select an agent, then open the **Persona** tab. The Instructions box sits at the top and Guardrails sit underneath it.

## Instructions

Instructions are free text. Most teams open with one line describing the character, then add a list of rules underneath. The toggle above the box switches between **Formatted** and **Raw**, so you can write it either way.

Things that belong in Instructions:

* The character the agent is playing sets everything else, for example an experienced host of a boutique short stay property with years of hosting experience.
* The way it addresses people matters, such as using the guest's first name from the reservation.
* Tone is worth stating plainly, for example warm, easygoing and conversational rather than corporate.
* Pronouns are a common rule, such as using "we", "our" and "us" and never "I", "my" or "me".
* Length and formatting keep replies readable, for example short sentences and a maximum of two or three sentences where possible.
* Emoji and sign off rules stop the agent sounding repetitive, for example varying the closing line and using a single unicode emoji rather than ascii emoticons.
* Disclosure rules belong here too, such as never claiming to be an AI or a bot if that is your policy.

Write each rule so you could tell whether the agent followed it. "Be friendly" is hard to check. "Keep responses to two or three sentences" is not.

## What belongs somewhere else

* Facts such as check-in times, policies and property details belong in **Knowledge**.
* Step by step processes and anything conditional belong in **Skills**.
* Tone that only applies to one scenario belongs in the Skill for that scenario, not in the Persona.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.