Support inbox

Support inbox and human hand-off: the agent knows when to stop

A support app in AgentRow gives its agent one more tool: hand this conversation to a person. The moment it is called, the conversation moves to your team's inbox, the customer is told, and your reply reaches them in the same chat, under your name.

What is an AI to human hand-off?

A hand-off is the moment an AI assistant stops answering and a person takes over the same conversation. It is what separates a chatbot that deflects from a support product: the customer asked for a person, a decision needs a human, or the assistant could not answer, and the thread continues with someone who can.

In AgentRow the hand-off is a tool the agent can call, not a rule hidden in a prompt. You decide when it should hand off and how it should reply when it does, in plain language, per app.

How does the hand-off work in AgentRow?

Make the app a support app and attach the Support hand-off node to its Agent node on the Workflow tab. The node names the team and carries two prompts: when to hand off and how to reply after handing off. The agent gets a built-in hand-off tool with those instructions.

When the agent calls it, the run completes with the agent's own hand-over message, the conversation moves to the team, and the widget shows the customer that a person will reply. From then on the customer's messages are stored for the inbox and the AI does not run again on that thread until it is resolved.

What does the team see in the inbox?

Inboxes lists every published app. A support app's inbox has Open and Resolved tabs; each conversation opens on the right with the full history, the agent's reason and summary for the hand-off, and a reply box. Replies reach the customer's chat with the team member's name.

Resolving closes the thread; a new message from the customer reopens it. Apps that are not support apps still get a read-only inbox, so every conversation is visible in one place.

What is on the record, and what comes next?

The hand-off shows in the run's logs like any other tool call: the reason, the summary, the team, and whether the customer got the agent's own words or the configured hand-over message. There is no silent escalation and no rule outside the trace.

Slack threads already land in the same inbox. Coming in the next months: support roles and seats so team members see only the inbox, email notifications on every hand-off, AI-drafted replies your team can edit before sending, replies delivered back into Slack, and WhatsApp.

Frequently asked questions

Does a hand-off cost extra per resolution?

No. A hand-off is part of the conversation it belongs to. Team members who reply from the inbox are priced as support seats; the AI is not priced per outcome.

Can I make the agent say exactly the same thing on every hand-off?

Yes. The node's reply prompt is given to the agent as written, so 'Reply with exactly: …' produces the same sentence every time, while a looser prompt lets it explain in its own words.

What if the agent hands off when it should not?

Tighten the when-to-hand-off prompt. The reason it gives is in the trace, so you can see what it was reacting to. A hand-off never fails the run.

Build it in AgentRow

Start a free trial, or read how the rest of the platform fits together.