Zurück zum Glossar

Function Calling: Definition, How it Works & Use Cases

Function Calling is a mechanism where an AI system calls external functions or services at runtime, rather than deriving results solely from its internal knowledge. This allows the model not only to generate text but also to perform specific actions and interact with other components of an IT system. In practice, this approach is also referred to as "Tool Use" – the model utilizes "tools" in the form of functions or APIs.

‍

What is Function Calling?

Function Calling describes the coupling of an AI model with external functions via Application Programming Interfaces (APIs). The model recognizes when a user request or workflow requires an external action. Instead of simulating the desired information itself, it selects an appropriate function and provides the necessary inputs. The result – data, a confirmation, or an error message – then feeds into the model's response or the next processing step.

‍

How Does Function Calling Work?

Technically, Function Calling is based on APIs, which serve as a set of rules for exchanging requests and responses between system components. An orchestration or mediation layer decides, based on the context, which function is responsible. Developers define the conditions under which functions are executed, as well as the required parameters and output formats.

‍

A typical process involves five steps:

     
  1. Recognition – The model determines that an external function is needed (e.g., for current weather data instead of pure text generation).
    2.  
  2. Selection – The appropriate function is identified.
    3.  
  3. Parameterization – The necessary input values are compiled.
    4.  
  4. Execution – The external function or service is called.
    5.  
  5. Integration – The return value is integrated into the model response or used as the basis for the next step.

‍

Depending on the implementation, an error may occur. The system can check the validity of the results and verify the output before further processing. The modular division into separate components allows the process to be implemented as a structured workflow automation, which increases reliability for multi-step tasks.

‍

Practical Examples and Use Cases

Function Calling covers a wide range of business and IT processes.

     
  • Customer Support: Resetting passwords, scheduling calls, updating customer data.
    •  
  • E-Commerce: Checking stock levels and availability in real-time, processing payments via external services.
    •  
  • Smart Home: Controlling devices such as lights, thermostats, or security systems.
    •  
  • Financial and Compliance Processes: Verifying transactions and issuing compliance alerts for unusual activities.
    •  
  • Medical Triage: Diagnostic and prioritization support in clinical workflows.

‍

Distinction: Function Calling vs. RAG and LLM Agents

Function Calling is often compared with two related concepts.

‍

Retrieval-Augmented Generation (RAG) primarily retrieves information from a knowledge base or documents via search and uses it for answer generation. Function Calling, on the other hand, calls a specific function or API to directly execute a defined task and obtain a real-time result.

‍

LLM Agents control a sequence of steps with the help of a language model and proactively decide on actions. Function Calling, in contrast, specifically addresses the precise execution of defined tasks via external tools within a fixed workflow.

‍

Conclusion

Function Calling connects AI models to external functions via APIs, enabling the controlled, condition-based execution of actions. The results are directly integrated into responses or subsequent steps. This clearly distinguishes the approach from RAG, which relies on document retrieval, and from agents, which independently plan sequences of actions. For workflows requiring real-time data or system-level operations, Function Calling is a precise control instrument.