Use case
Support answers without a database login for everyone
Support teams spend their day looking things up in systems they were never trained on. Connect the database or API those answers live in, and an agent finds the customer, the order and the ticket — while anything that changes money or deletes data waits for a person.
In practice
What support asks
- “What is the status of order 4471?”
- “Find the customer with email jane@example.com and list their last five orders.”
- “Has this customer been refunded before?”
- “Open a ticket for this customer about a late delivery.”
What keeps it safe
- ✓ Destructive operations need approval by default
- ✓ An approval covers one call, with the arguments a person reviewed
- ✓ In web chat, each person acts under their own role’s permissions
- ✓ Every change an agent makes is recorded in the change log
How it works
From connection to answer
-
01
Connect the systems
Your database, a REST or OpenAPI service, or an MCP server — whichever holds customers, orders and tickets.
-
02
Review the drafted agent
Caterfli drafts an agent from what it discovered. Turn off what support should not touch, and decide what needs approval.
-
03
Answer from where the team works
Web chat at their desks, or Slack, Telegram and WhatsApp for the people you allow.
Works with
Read the documentation
Can a support agent issue refunds?
Only if the connected system exposes a refund operation and you leave that tool enabled. You choose per tool whether a call waits for approval, and an approval covers exactly one call with the arguments a person reviewed.
Do support staff need database access?
No. They talk to the agent, and the agent uses a connection whose credential is encrypted and never shown to the model, to the browser, or to them.
What happens when an approval is needed?
The run pauses and the request appears in the approval queue with the tool, the connection and the exact arguments. Once a person approves, that one call runs; if the agent re-plans, it has to ask again.
Try it on your own system
Connect one system and read the agent Caterfli writes from it. There is a demo connection too, if you would rather look before you wire anything up.