Яндекс.Метрика

Tool calling

Tool calling explained

A language model's ability to use external functions like APIs, search and databases. It's what lets an AI agent take action, not just answer

Tool calling (also called function calling) is how a language model works with external systems. The model gets a list of tools with descriptions and parameters. When a task needs a tool, the model returns a call request instead of text: which function to run and with which arguments. The program runs the call and hands the result back to the model, which decides what to do next.

Examples of tools:

  • Send a message or read a thread's history
  • Create a task in a tracker
  • Read a file in a repo
  • Search the web or a knowledge base

Tool calling is what sets an AI agent apart from a chatbot that only replies. The MCP protocol helps connect tools in a common format.

An agent works with Pachca through the API: it sends messages, reads threads and creates tasks and reminders. To help the agent call methods correctly, there's Pachca CLI with command details via --help, Agent Skills and the OpenAPI spec.

Learn more in the Help center

Read the article →

Related terms