Skip to main content

How the AI agent works

For every customer message on a contact the AI owns, Loomi builds a reply from:

  1. Your instructions — tone, length, personality, things to avoid, custom rules.
  2. The knowledge base — the most relevant chunks of your documents, pages and Q&A for this message, plus static sources that are always included.
  3. The catalog — matching products from Shopify or WooCommerce, with live price and stock when it quotes them.
  4. The conversation — the recent history, attachments described as text, the ad the customer came from, open tickets and the contact's details.

What it can do​

  • Answer questions from the knowledge base and say when it does not know.
  • Search products, check inventory, and describe items with links.
  • Collect name, phone and email (email verified by code), and save them on the contact.
  • Book and manage appointments when a calendar is connected.
  • Open support tickets and flag them for a human.
  • Hand over to the team, routed to the right department.

Rules it always follows​

  • Replies in the customer's language, from the latest message. Never Hebrew.
  • Keeps every reply under 1,000 characters.
  • Never invents details a source does not contain, and never retracts an earlier reply on its own.
  • Answers a burst of several short messages once, not once per message.
  • Treats a generic ad as not naming a product and asks which item the customer means.

After each reply​

A classifier tags the contact's intent, sentiment and segment (hot lead, warm lead, support seeker…) and may move them along the pipeline. Lead labels are only given for real interest in something you sell.