Skip to main content
You don’t call these tools yourself. You ask in plain words and the AI app picks the tool. This reference is for knowing what is possible and what each tool returns.

​
All tools

​
Tool hints

Each tool tells the AI app how careful to be:
  • Read-only tools only look. AI apps usually run them without asking.
  • Destructive tools (update_agent, connect_phone_number) overwrite live settings with no undo. AI apps ask you to confirm first.
  • Open world tools (the three create tools and retry_agent_build) reach outside VoiceAIWrapper: the build reads your business’s website.

​
What a tool returns

Every reply has:
  • Text written for you, which the AI app usually shows or summarizes.
  • Structured data (structuredContent) that follows the tool’s published output schema. The same data is also sent as a JSON text block, for AI apps that only read text.
A refusal comes back as a tool error with a sentence saying what happened and what to do next. See Errors.

​
Ids

Tools hand out and take opaque ids:
  • agent_id: an agent, from list_agents or a finished build. It is the same id the VoiceAIWrapper app uses, so a link from the chat opens the right agent.
  • build_id: one agent build, from a create tool.
Ids only work in the connected organization.

​
Repeating a request safely

The create tools and create_web_widget take an optional request_id. Send the same request_id when you retry the same request and nothing is built twice:
  • Same request_id, same details: you get the first result back.
  • Same request_id, different details: refused. Use a new request_id for a new request.
  • No request_id, same details within 10 minutes: treated as the same request.
Request ids are kept for 24 hours.