Give the assistant tools to search your products and open a ticket, order, or booking, and set when it hands the conversation to a human agent.
Tools are the capabilities that let the assistant interact with your apps instead of just talking: searching your product catalog, fetching an order's status, opening a support ticket, registering an interested lead, or booking an appointment. With them, the assistant answers with live, accurate information and takes a real action on your behalf. Tools are managed from the Basics tab on the AI Chatbot settings page.
The counter above the grid shows how many tools are enabled out of the total.
Tools fall into categories that the system marks with a clear badge:
| Type | What it means |
|---|---|
| Read tool | Only fetches information (search, lookup). Safe, and included in "Enable all." |
| Write tool 🖊 | Creates or changes real data (opening an order, registering a lead). Sensitive; stays off by default, and each must be turned on manually and deliberately. |
| Always on | Foundational tools that can't be turned off. |
| Workflow | Tied to an automation flow and managed from that flow's page, not here. |
Important note: when you click Enable all, only read tools are turned on. Write tools are sensitive and keep their current state — enable each one deliberately when you need it.
Which tools are available depends on the apps you have installed. Examples:
| App | What the assistant can do |
|---|---|
| Store & Orders | Search products and state price and availability, and create an order. |
| Car Dealership | Search the listed vehicles and state their specs, and register an interested lead. |
| Real Estate | Search available properties, and register a lead linked to the right property. |
| Booking | Show available slots and book an appointment for the customer. |
| Support Tickets | Open a support ticket from the conversation automatically. |
The practical rule: enable read tools (like search) without hesitation, and enable write tools (like creating an order or registering a lead) after you've tested them and are comfortable with the assistant's behavior in the Playground.

However good the assistant is, some conversations should be handled by a person: a sensitive complaint, a complex request, or an upset customer. The system provides three layers for a safe handoff, configured in the Guardrails tab.
Words that, if the customer types them, immediately transfer the conversation to an agent. Add words like "agent," "talk to a person," or "complaint." Type the word and click add to see it appear as a removable tag.
Sensitive words or situations that make the assistant escalate automatically instead of trying to answer (for example: "cancel subscription," "legal," "refund").
The ready-made instructions include an explicit rule: if a question falls outside the company's scope, or the problem isn't resolved after two or three replies, the assistant calls the human-handoff tool instead of inventing a solution. You can reinforce this behavior in the instruction text in the Persona tab.
We use essential cookies to run the site. With your permission we also measure usage and ad performance. You can change this anytime. Privacy policy