Your AI Agent handles most conversations on its own. But some contacts want a real person, and when that happens you want the agent to step aside and make sure your team knows to pick up. Human intervention is how the agent detects those moments and pauses for you.

To get there, click AI Agent in the sidebar, then Human intervention.

The Human intervention panel in the AI Agent settings.

Human intervention vs. handing off to the AI Agent

These two sound alike but do opposite things, so it's worth being clear:

Human intervention (this article) is when the agent hands off to a person. It detects that a contact wants human help, then pauses so your team can take over.
Hand over to AI Agent is a scenario action that hands the conversation to the AI Agent, not a teammate. It's how a flow passes an open-ended chat to the agent to handle. That one is covered in AI Smart Actions in Scenarios and Agent Handover: Task Completion & Expiration Branches.

If you want a human to take the conversation, you're in the right place.

How detection works

When you turn human intervention on, the agent watches each message for signs the contact wants a person. It picks up direct requests ("Can I speak to someone?") and indirect ones (clear frustration, or a question it couldn't answer after a few tries).

When it spots one, the agent sends a short acknowledgement, then stops replying and waits for a human. The conversation is marked as needing a person, which pauses the agent for that contact until your team steps in.

Setting it up

Open the Human intervention panel and turn on the toggle. A notice reminds you that enabling it pauses the AI agent until a team member replies to the contact.

The Human intervention setup with the toggle and the 'trigger this automation' scenario picker.

Then pick the scenario to run when a request is detected, under If contact asks to speak to a human, trigger this automation. Selecting a scenario is all this panel asks of you, and the scenario is where everything else happens. Inside it you can:

Notify your team by email or a Slack message.
Add the contact to a "Needs human" folder so you can filter the inbox to everyone waiting.
Send a holding message that sets expectations.

So if you want an email alert or a message sent the moment someone asks for a person, build those steps into that scenario.

What the contact sees

When a request is detected, the agent sends a short acknowledgement that a team member will follow up, then goes quiet for that conversation. It keeps this brief, something like "Got it, a member of our team will get back to you shortly." The contact doesn't need an explanation.

If you'd rather send your own wording, add a holding message as a step in the scenario you connected above.

Taking over a conversation manually

You can step into any conversation from the Inrō inbox at any time, without waiting for the agent to detect a request. Open the conversation and reply, and the agent stops replying to that contact.

When the agent is actively handling a conversation, the inbox shows an "agent handled" banner with a Remove Agent from conversation button. Click it to take the conversation back by hand. Either way, the agent won't talk over you. For more on who's in control of a conversation, see Handling Status: Human, Automation, Agent & More.

⚠️ If your activation rules are set to Disable all replies, the agent won't reply at all, including the acknowledgement on a human request. Make sure your activation rules let the agent respond before relying on this.

🐾 Netsuke's Tips

Wire the request to a scenario that emails or Slacks your team. Relying on someone happening to check the inbox means contacts sometimes wait too long.
Review the conversations that triggered human intervention each week. If the agent is escalating things it should have handled, your knowledge base probably has a gap.
If a lot of requests cluster around one topic, add a knowledge entry on it. The agent will handle it next time instead of escalating.
If your scenario sends a holding message, don't promise a human will reply "immediately." "Shortly" or "as soon as we can" sets a more honest expectation and causes less frustration.