Why should you care?

PromptVerse agents declare functions like search_docs and create_ticket for the model to invoke.

See it live — copy this example

Copy the prompt into ChatGPT, Claude, or your LLM API playground and compare outputs.

tools = [{ type: "function", function: {
  name: "search_docs",
  parameters: { type: "object", properties: { query: { type: "string" } }, required: ["query"] }
}}]
messages = [{ role: "user", content: "Find refund policy for annual plans" }]
// model may return tool_calls[0].function.arguments

What happened?

• Tools array defines callable surface.
• Model emits tool_calls when it needs data; your server runs search_docs and returns tool role message.

Practice next

1. Register one function in playground.
2. Trigger a query needing search.
3. Execute function server-side.
4. Add optional top_k integer param.
5. Require human approval before destructive tools.

Remember

Schema defines safe parameters. Server executes — model does not. Allow-list tool names.

Doc search tool

Chatbot answers policy questions.

Outcome: Model calls search_docs instead of guessing policy text.