Tool use (function calling)

Tool use (sometimes called "function calling") is the LLM capability that enables AI agents. The model is given a list of available tools (functions), each with a name + description + input schema. When responding, the model can decide to call one or more of these tools with structured arguments rather than just producing text.

This is what turns a "chat model" into an "agent." Without tool use, an LLM can only produce text. With tool use, it can take real actions in the world.

Example tool definitions an agent might have:

• search_crm(email) — search the company CRM by email, returns customer record
• send_email(to, subject, body) — send an email via Resend
• book_meeting(attendees, duration) — book a meeting via Cal.com
• query_database(sql) — execute a SQL query against the analytics DB

The agent loop runs: model receives the user request + tool list → model decides which tool to call + with what arguments → system executes the tool → result returned to the model → model decides next action.

Modern LLMs (Claude 4.x, GPT-4o+, Gemini 2) all support tool use natively. The major standard is now MCP (Model Context Protocol) which defines how tools should describe themselves so any model can use them.