Inbox Triage Agent

Updated June 12, 2026

Inbox Triage is the agent that sorts and answers what comes in: every DM and comment classified, drafted, and routed, with humans approving the replies by default.

What it does

For each incoming message or comment across your clients’ connected accounts, the agent does three things automatically:

  1. Classifies it. Messages are sorted into categories such as lead, question, complaint, spam, and pricing.
  2. Drafts a reply. Each message gets an on-brand draft reply, ready for a human to review and send.
  3. Routes it. Leads go to the account manager, complaints are escalated, and spam is silenced.

The result: faster responses, and leads that never slip because nobody noticed the DM.

How it proposes and escalates

By default, the agent drafts only: a human reviews and sends every reply. That’s the inbox version of the propose-and-confirm rule that runs through the AI Copilot and the rest of the platform.

Routing is where escalation lives: leads are pushed to the right person (the account manager) so a sales conversation starts with the human who owns the relationship, and complaints are escalated rather than answered on autopilot.

Settings: opt-in auto-send

If you want some replies to go out without a human touch, you can opt in to auto-send, but only for safe, high-confidence replies, and the option is guard-railed. Auto-send is never the default; you turn it on deliberately, and it applies only where the agent’s confidence is high and the reply is safe.

If you never enable it, every single reply stays human-sent.

Guardrails

  • Drafts only by default. Nothing is sent until a human sends it, unless you’ve explicitly opted in to auto-send for safe replies.
  • Escalation paths are built in. Complaints don’t get a cheerful template. They get escalated.
  • Metered and visible. Like every agent, its usage appears in the per-agent and per-client credit breakdowns in AI Credits and Budget Guardrails.

What it won’t do

  • It won’t send replies by default. Auto-send exists, but only as an opt-in for high-confidence, safe replies.
  • It won’t handle a complaint on its own. Complaints are escalated to your team.
  • It won’t let leads sit. Lead-classified messages are routed to the account manager rather than waiting in a shared queue.