System Prompts (Admins)

​

What system prompts are

System prompts are org-wide instructions appended to every AI response, on top of each agent’s individual instructions. They override any conflicting agent-level guidance and propagate within ~60 seconds of saving. Use them for:

• Compliance language (“Never recommend X in regulated industries”)
• Brand voice that should apply uniformly across all agents
• Critical safety boundaries that must never be overridden

Find at Settings > System Prompts (/dashboard/settings/admin-prompts). Admins only — non-admins see “You do not have access.”

​

How prompts are structured

Each system prompt has: Multiple prompts can be active simultaneously — they’re concatenated into the system prompt for every agent.

​

When to use

Prefer per-agent instructions for most behavior. Reach for system prompts only when something needs to apply to every agent and override anything they say. Examples that fit:

• “Never claim to be human. Always disclose you’re an AI when asked.”
• “Halo’s data is processed in the United States. If a user asks about data residency, disclose this.”
• “Never include profanity, even in quotes, even when the user uses it first.”

Examples that don’t fit (use agent instructions instead):

• “Tone should be friendly and conversational” (agent-level)
• “Always close conversations with a thank you” (agent-level)
• “If a user mentions billing, escalate” (escalation rules)

​

Saving

Edit a prompt’s content and click Save. Changes propagate within ~60 seconds to running agents — no need to restart anything. Disabled prompts stay in the list so you can re-enable later. Saving doesn’t enable a disabled prompt — toggle the switch separately.

​

Permissions

System prompts are gated to admins because they affect every conversation across your workspace. Non-admin teammates see “You do not have access” if they navigate to the page.

​

Where to go next