It answers from your systems, not just from your documents
Uploading a PDF gets you an AI that knows your opening hours. Connecting a database gets you one that knows this caller's balance, whether that part is in stock, and which slot is genuinely free on Thursday.
Read-only, always. Every question the AI asks of your data goes through the same guard as every other query — no writes, no schema changes, nothing that could alter a customer's record. Different call links can be given different slices, so the sales line and the support line don't see the same thing.
Twenty-five types. If yours isn't listed, ask — the connector layer is the part designed to be extended.
And the systems you don't run yourself
Most businesses keep more in other people's software than in their own database — the CRM, the helpdesk, the payment processor, the shop. There is a wider catalogue of connectors covering those: CRM, support, payments, e-commerce, email, analytics, storage, social and automation.
They reach the AI differently from a database, and the difference is worth understanding before you buy rather than after.
Asked at the moment it matters
The AI forms a question while the customer is talking and gets the answer back inside the conversation. Read-only, guarded, and true at that second.
"Your balance is £340 and the payment cleared on the 3rd."
Gathered first, then handed over
Your connected systems are read before the conversation starts and the picture goes in with the brief — revenue from the payment processor, contacts from the CRM, open tickets from the helpdesk, orders from the shop.
The AI knows the shape of your business and which systems it can be asked about. What it cannot yet do is call one of those APIs during a sentence, the way it can a database.
"I can see you've an open ticket with our support team about that."
Live data pulled today from Stripe, HubSpot, Salesforce, Zendesk, Intercom, Shopify, Google Analytics and Meta Ads. The wider catalogue is connectable and reportable; if you need a specific one answering questions live on a call, ask us where it sits on the list rather than assuming.
Real figures, not approximations
"Your balance is £340 and the payment went out on the 3rd" rather than "let me take a message and someone will check". The difference between a call that resolves and a call that creates another call.
The list is a query
Everyone overdue, everyone due a recall, everyone booked tomorrow. It runs again each time the campaign does, so the list is never yesterday's.
It cannot change anything
Every query is checked before it runs. No writes, no deletes, no schema changes — a campaign that could run arbitrary commands against a customer's production database is a campaign that can destroy it.
Per link, not per company
Point one link at the parts of your data that answer sales questions and another at the parts that answer support questions. Deleting a scope denies access rather than quietly widening it.