Tools — what does the AI assistant use behind the scenes?
When the AI assistant works, it uses a range of tools to find information and perform actions. You see them activate in real time as status messages in the panel — e.g. "Searching help center..." or "Looking up order...".
Tools are divided into three categories:
Mandatory tools (always run)
| search_help_center | Searches the Guide help center for relevant articles |
| search_macros | Searches Zendesk macros for relevant templates |
| search_accepted_responses | Searches previously accepted AI responses for similar cases |
| get_instructions | Retrieves admin-configured instructions and conditional rules |
If all mandatory tools return empty results, the AI assistant does not generate a draft — instead it asks the agent a question.
Situational tools (run when relevant)
| search_ticket_history | Searches the customer's previous tickets for similar cases and solutions |
| get_ticket_fields | Retrieves ticket fields and metadata |
| manage_ticket_field | Updates ticket fields (type, priority, etc.) based on analysis |
| read_attachment | Reads text content from attachments (OCR) |
On-demand tools (on request)
These tools are only activated when the agent asks for it — via quick action buttons or directly in the chat field:
| web_search | Searches the web via Brave Search (🌐 Web search button). Requires activation in admin. |
| fetch_url | Retrieves content from an allowed URL (configured by the admin under Allowed URLs) |
| magento_get_order | Looks up a Magento order by order number |
| magento_get_orders_by_email | Looks up Magento orders by email |
| manage_ticket_tags | Adds or removes tags on the ticket |
| list_side_conversations | Lists side conversations on the ticket |
| create_side_conversation | Creates a new side conversation |
| save_rule | Saves a rule from the chat (enabled via the save-rules feature) |
| list_rules | Lists active rules from the chat |
HTTP tools configured by the admin under Integrations also appear as on-demand tools.
What happens when a tool does not find anything?
If all mandatory tools return empty results, the AI assistant does not generate a draft. Instead, it asks the agent a concrete question — for example:
"I couldn't find information about your returns policy. Can you tell me what it is, so I can write a response?"
This is intentional: the AI assistant does not invent information. It asks for help rather than guessing.