An AI inbox is not a chatbot attached to WhatsApp. It is a communication layer that classifies, summarises and prepares actions on the channels a company already uses. The perimeter has to be written down. Human supervision is not an extra: it belongs in the design.

What an AI inbox is meant to fix

The usual friction is not a lack of channels. It is that WhatsApp, email, forms and internal threads arrive without a shared priority, a shared history or a stable rule for what gets answered, escalated or closed.

Onezix Inbox is an operating inbox: it reduces noise, gathers context and leaves exceptions ready for the person who must decide. It does not replace the team. It organises work the team already does.

Set the perimeter before connecting channels

Perimeter means deciding, in writing, which conversations the AI layer will handle, under which identity it replies and what stays out. Without that edge, support, sales, internal incidents and personal messages collapse into one queue.

The perimeter is not improvised on the first thread. It is agreed with the people who run the channel: hours, languages, request types, data that must never leave the system and accounts that are not automated.

  • Included channels and channels that are explicitly out of scope.
  • Visible identity: brand, team or a specific mailbox.
  • Conversation types the AI may draft versus types it only labels.
  • Data that is not summarised, forwarded or copied into other tools.

WhatsApp and email are not operated the same way

WhatsApp concentrates immediate waiting, short messages and changing interlocutors. Email concentrates long threads, attachments and people in copy. Treating both with one template produces the wrong tone or the wrong timing.

An inbox layer should respect each channel’s pace: WhatsApp often needs classification and a short draft; email needs the thread, the attachments and who is copied. Escalation rules change with that pace.

Human supervision: when a person steps in

Supervision is not reading every message twice. It is defining the points where AI does not close on its own: a complaint, sensitive data, a client with an open incident, a commercial commitment or any case without a clear rule.

The team needs the reason for the escalation, the context already gathered and the proposed action. If a person has to rebuild the thread from scratch, the AI inbox is not operating: it is adding a step.

Draft, send and record

In production it is useful to separate three acts: classify, draft and send. Many companies start well by classifying and leaving drafts. Sending in the company’s name needs a threshold, a permission and a trail.

Every relevant action should be recorded: what was proposed, who approved it, what was sent and which system was updated afterwards. Without that trail the inbox cannot be audited or improved.

How Onezix OS deploys it

Onezix Systems deploys Onezix Inbox on existing channels, not as an app where the team signs up and starts alone. The design includes perimeter, permissions, priority rules and supervision points agreed with operations.

Connected to Knowledge and Workflows, the inbox does more than reply: it consults internal sources and leaves actions ready in CRM, tasks or reporting. A messaging connector moves messages. Inbox, Knowledge and Workflows share the case, the source and the next step.