Function Calling

Function calling (also known as tool use) is a capability built into modern language models that allows them to interact with external systems. Instead of only generating text, the model can output a structured request to call a specific function — such as querying a database, sending an email, checking the weather, or executing code. The model does not execute functions directly. Instead, it generates a function call request (including the function name and parameters) in a structured format. The application layer executes the function and returns the result to the model, which then incorporates the result into its response. This creates a powerful loop where the model can reason about what tools to use, request their execution, and interpret the results.

Developers define a set of available functions with descriptions and parameter schemas. When the model receives a user query, it decides whether any available function would help answer it. If so, it generates a structured JSON call with the appropriate function name and arguments. The application code receives this call, validates it, executes the corresponding function, and passes the result back to the model. The model then uses the result to formulate its final response. Some queries may require multiple sequential function calls — for example, first looking up a customer, then checking their order status, then generating a response. Modern models are trained to make intelligent decisions about when to call functions, what parameters to provide, and how to interpret results. They can also handle parallel function calls when multiple independent pieces of information are needed simultaneously.

Function calling transforms language models from conversational tools into action-taking systems. It bridges the gap between AI understanding a request and AI fulfilling it. Without function calling, an AI assistant can tell you it would be helpful to check the database — with function calling, it actually checks the database. This capability is the foundation of AI agents, automated workflows, and intelligent integrations. It allows organisations to connect AI models to their existing systems — CRMs, ERPs, databases, APIs, communication tools — creating AI-powered interfaces to their entire technology stack. Function calling also improves accuracy by giving models access to real-time data. Instead of relying on potentially outdated training data, the model can retrieve current information from authoritative sources, combining the reasoning capabilities of AI with the accuracy of live data.

Function calling powers a wide range of applications. Customer service agents use it to look up orders, process refunds, and update records. Data analysis assistants use it to query databases and generate visualisations. Scheduling assistants use it to check calendars and book meetings. In software development, function calling enables AI to execute code, run tests, and interact with version control systems. In business operations, it connects AI to workflow automation platforms, enabling end-to-end process automation. Any scenario where an AI system needs to interact with external services relies on function calling as the underlying mechanism.