docs / human-handoff-and-guardrails

Human handoff & guardrails

The guardrails aren't a settings page you have to go find — they're on by default. Here's exactly how Miira stays on a leash, and how you take a conversation back.

always disclosed

It never pretends to be human.

// Every thread tells the lead it's talking to an AI.

Miira identifies itself as an AI agent in the conversation. There's no setting to make it pose as a person — disclosure isn't optional.

instant handoff

Take the chat, and the agent goes quiet.

// Mid-conversation, no delay.

When you "take over the chat," Miira goes silent immediately — even in the middle of a reply. From that point the conversation is yours; the agent won't talk over you.

autopilot is off

It starts off, with a kill switch.

// You turn it on deliberately — it's never on by surprise.

  • Autopilot starts OFF. The agent doesn't send on its own until you decide it should
  • There's a kill switch to stop it at any time
  • It only ever sends over Meta's official WhatsApp Cloud API — never an unofficial channel

never alone

It never moves your pipeline on its own.

// The agent suggests. A human decides.

Miira suggests moving a lead to qualified — it never moves a client's stage without a human confirming it. And a client's stage only ever changes through a legal transition, never a free-text field, so the record can't drift into a state that doesn't make sense.

Closing a client or losing a deal is a soft delete with a full audit trail — never a hard erase. Nothing you turn off is truly gone.

What this guide doesn’t cover

One thing to be clear about:

  • These aren't features you have to enable — they're on by default, and most can't be turned off by design.
  • Uptime, support hours, or how long any AI provider retains data are not guaranteed here beyond what's stated in our Privacy Policy.

Last updated: 2026-07-15.

← All guides · See the roadmap