Activation rules decide when your AI Agent jumps into a conversation and when it stays quiet. You can have it answer every incoming message, step in only for the topics you pick, sit in the background until an automation calls on it, or go silent altogether. This article covers every option.
Go to AI Agent → Activation rules to set this up.
The four activation modes
You pick one mode, and only one can be active at a time. New agents start in Specific intents only, so the agent does nothing until you give it a topic to watch for.
All pending messages
The agent replies to every incoming DM that a running scenario hasn't already handled. This is the right choice when you want the agent to be your main responder for all conversations.
If a scenario answers a message first, the agent stays out of it. Scenarios always take priority over the agent, as covered further down.
Specific intents only
The agent replies only when it detects that an incoming message matches one of the intents you've defined. An intent is a plain-language description of what a contact wants. Everything else is left alone.
When you choose this mode, an AI-detected intents field appears. Add your intents one at a time, phrased the way a real person would ask:
"asks for a refund"
"wants to book a meeting"
"asking about pricing"
"has a question about an order"
The agent checks each message against your list. If any one intent matches, the agent takes over. Matching is done by AI, so it catches paraphrases, not only exact keywords.
Once the agent picks up a conversation this way, it keeps replying to follow-ups in that thread without re-checking intents every time. The thread returns to its previous state after 24 hours with no activity, and the agent then needs a fresh intent match to re-engage.
⚠️ If you leave the intents field empty in this mode, the agent replies to nothing. That is handy while you set things up before going live.
Automation only
The agent never replies on its own. It acts only when a scenario hands the conversation to it, using an AI action like Hand over to AI Agent, AI Agent message, or AI Agent auto-DM.
Choose this when you want full control through your flows, with the agent stepping in only at moments you've designed. See AI Smart Actions in Scenarios for how to call the agent from a scenario.
Disable all replies
The agent goes completely silent. It won't send anything, including messages triggered from inside a scenario's AI actions. The option is shown in red as a reminder that it switches the agent off entirely.
Use this to pause the agent without losing your setup. Your personality, knowledge base, intents, and actions all stay saved for when you turn it back on.
⚠️ If your account runs on free AI credits and they run out, Inrō switches the agent into this mode for you, so it stops replying until you upgrade. See AI Credits for how that works and how to keep the agent running.
Limit replies based on folders
Below the mode selector is a folder filter. It works on top of whichever mode you picked, so a message has to pass both checks before the agent replies.
Don't limit replies based on folders. The default. The agent applies its mode to everyone, whatever folders they're in.
Only reply to contacts in these folders. The agent activates only for contacts in the folders you pick. Everyone else is ignored, even if their message would normally get a reply.
Never reply to contacts in these folders. The agent works normally for everyone except contacts in the folders you choose. Good for excluding people you're handling by hand, VIPs, or a particular funnel stage.
⚠️ Folder rules apply to the agent's own replies only. They do not block the AI actions inside your scenarios: you can still send agent messages to a contact in an excluded folder from an automation flow.
How the mode and folder checks work together
The mode and the folder rule both have to pass before the agent replies. Picture two gates:
Does the message match the active mode? (a pending message, or the right intent)
Is the contact in an included folder, or out of an excluded one?
If either gate fails, the agent stays quiet.
When the agent stays quiet anyway
Even in All pending messages mode, the agent holds back in a few situations, by design, so it never talks over you or a flow:
A scenario already replied (or is about to). Scenarios take priority, and the agent fills the gaps and follow-ups around them.
A human took over. If a contact asks for a person, or a teammate steps in, the agent pauses for that conversation. See Human Intervention: Handing Off to a Human.
The contact already got an answer. If a newer message was sent or received after the one that triggered the agent, it won't pile on a late reply.
The message has no text. A bare story mention, photo, or video with no words won't get a reply.
Saving your changes
Click Save after any change. New settings take effect right away for every future incoming message.
🐾 Netsuke's Tips
Start in Specific intents only when you first go live. It keeps the agent focused on one or two conversation types, so you can watch it handle them well before opening it up to every message.
Build a "human-handled" folder and pair it with Never reply to contacts in these folders. When you take a conversation over by hand, drop the contact in that folder and the agent backs off on its own.
Write intents the way a customer would phrase the request, not as keywords. "Asks how much it costs" catches more real messages than "pricing inquiry".
Before switching from Specific intents only to All pending messages, check your knowledge base. The agent will now field everything, so any gaps in what you've taught it will show up fast.
What's next?
Now that the agent knows when to reply, give it something to reply with. Go to Knowledge Base: Teaching Your AI Agent What to Know to add documents, web pages, and notes it can draw on. To call the agent from inside a flow, see AI Smart Actions in Scenarios.