Quiet Core

Quiet Core is the assistant a sales agent opens on the phone before walking in to a customer. They type the customer’s name and get what matters: overdue amounts, the month’s sales, what is falling, what is rising, what needs following up and what can be proposed.

How it works

• The commands (/vizita, /briefing) are deterministic: the figures are calculated from the CRM, they are not generated by a model.
• It runs on an appliance installed at the company’s office; field data does not leave the company.
• The team’s notes are kept in the company’s memory, with access per person, team or company.
• Voice dictation for questions and notes.

The appliance

A computer at the company’s office. It runs the assistant, the local model and the link to the CRM. It is not exposed to the internet: the field team reaches it through a relay that stores no data.

We need one network port and read access to the CRM.

In more detail, from a real installation: What a server installed at the client looks like.

The brief, on the rep's phone

The rep types the command and the customer name. The answer comes from the CRM: overdue invoices, amounts within terms, sales this month and the change against the previous period. The customer name is blurred here, not in the app.

Notes on Quiet Core

• Field teams do not need open-ended chat — What an Area Sales Manager needs before a client visit, and why, instead of an open chat, we gave them fixed commands calculated from the CRM. Plus what we learned testing natural language conversations.
• What a server installed at the client looks like — What runs at the company's premises, how the field employee, the Area Sales Manager, reaches it from a phone, what goes to the cloud and who decides that. Plus the rule we apply: “Verified, not claimed.”
• The model is only one component — AI models change every few months. The data, the tools, the access rules and the test questions stay. Why these are the company's asset, and how we build around them.