What does the AI agent do, and what does it know?

The agent is a trained assistant that answers your customers — website enquiries, and WhatsApp where connected — in your name, from your data. It is configured under Workspace → AI agent.

What it answers from

The agent only knows what your workspace gives it:

• Your catalogue and price book — it identifies the exact vehicle a customer describes and quotes only prices that exist in your price list. It cannot invent a price: quoting runs through your pricing, not through the model's imagination.
• Your business facts — opening hours, services, policies you maintain on the agent page.
• Your uploaded knowledge — documents you add (FAQ sheets, warranty terms). Your uploads stay tenant-isolated: they answer for your workspace only and are preferred over general knowledge.
• Your calendar — when the booking calendar is on your plan, it can book an appointment in a free slot, and move or cancel one a customer already has, with the same standing on the calendar as you.

What it deliberately does not do

It does not discuss things outside your business, does not commit to work your price book cannot price, and does not keep talking when a conversation needs a human — it raises a handoff, which appears in your inbox with the reason, and the thread waits for you.

The Never say list under Knowledge is the hard floor: it ships with reviewed defaults, everything on it is refused in every conversation, and you can add your own lines to it.

The controls that matter

Everything the agent does is governed from Workspace → AI Agent, and each tab has its own article:

• Setup — enabling it, the connector key, and which channels it may answer on: How do I set up the AI agent?
• Templates — the WhatsApp messages approved for opening a conversation with a customer who has not written to you first: How do I connect WhatsApp?
• Knowledge — the documents it may quote (What documents should I give the agent?), the claims it may make actively (What can the agent promise on my behalf?), and the hard limits it refuses to cross (How do I stop the agent saying certain things?)
• Pricing — what it may say about money: What may the agent say about prices?
• Behaviour — how it sounds (How do I control how the agent sounds?) and whether it sends or drafts (How do I approve replies before they're sent?)
• Learning — reusing answers your colleagues already wrote: How does the agent learn from my team's replies?

Two switches are worth knowing without reading further. Disable agent on the Setup tab is the off switch — inbound enquiries still land in your inbox, nothing is answered until you enable it again. And a channel switched off means the agent never answers there, whatever else is configured.

Plan and limits

The agent is part of the Max plan, with no ceiling on conversations. If the workspace moves to a plan without it, the agent switches itself off on the next message and your inbox takes over; nothing that already came in is lost. Tuners who ran the agent before it became a Max feature keep it.