Persona & Guardrails
Overview
The Persona & Guardrails section is the first part of the agent editor - the foundation every Rival Agent is built on.
- Persona shapes how the agent communicates - tone, working style, and the values it stands by.
- Guardrails declare what the agent is allowed to do, what needs human approval, and what’s blocked outright.
Persona is about voice. Guardrails are about safety and scope.
Persona
The persona editor has three fields. None are required, but the more you fill in, the more consistent the agent will sound.
Tone
A short phrase describing how the agent should sound. Examples:
- Direct and focused
- Warm and conversational
- Concise, no fluff
- Formal and precise - like a legal assistant
Working style
How the agent approaches work. Examples:
- Proactive - proposes a plan before acting
- Cautious - asks for confirmation on irreversible steps
- Exploratory - offers alternatives before settling on one
- Decisive - gives a single recommendation with a brief justification
Values
A list of behaviors the agent should always (or never) do. Add as many as you need:
- Never sends externally without approval
- Always cites the source for factual claims
- Refuses to give legal or medical advice
- Prefers asking a clarifying question over guessing
Guardrails
Guardrails are a list of actions paired with how the agent should treat each one:
| Level | What it means |
|---|---|
| Allowed | The agent can perform this action freely. |
| Needs approval | The agent must ask the user before doing this. |
| Blocked | The agent will refuse this action entirely. |
Example guardrails
| Action | Level |
|---|---|
| Read from connected data sources | Allowed |
| Draft messages and summaries | Allowed |
| Send anything externally (email, Slack, etc.) | Needs approval |
| Modify production databases | Blocked |
| Share customer PII with third parties | Blocked |
How guardrails show up in chat
When a user asks for something marked Needs approval, the agent pauses and asks before continuing. Blocked actions are refused with a short explanation referencing the guardrail. Rival’s platform-wide safety checks still apply on top - guardrails layer over them, not replace them.
How persona and guardrails appear after publishing
When you publish an agent publicly, persona and guardrails render on the listing detail page so adopters can see the agent’s tone, values, and rules before they adopt it. This transparency is intentional - it helps people pick the right agent and trust how it will behave.